I’m testing something.
My plan is to build a universe step by step: write the novel, write the script, then film it. The problem? It’s a silo. Feedback arrives far too late. I can end up realizing an entire character doesn’t work only after spending a year writing them.

I’ve been inspired recently by the concept of Emergent Canon. Instead of a rigid plan, why not probe the audience in real-time? Float a character, a piece of lore, a design. Whatever spikes gets promoted to canon. Whatever dies, gets quietly buried.

So, I’m applying that to my sci-fi project. No more waiting. Let’s cook.

Subscribe now

Step 1: The Core Design
I started with a prompt blending high-fashion editorial with AI godhood. The goal was a “Spartan” robotic entity glowing with gold data.

Meet one of the Maximized Artificial Governing Intelligences.

The initial Midjourney output gave me a fantastic anime-style design. I then used GPT Image 2 to transpose that into a hyper-realistic live-action aesthetic and generated a secondary character—the “Engineer” who interprets her predictions.

The MAGI and the Engineer.

Step 2: The Storyboard (The 30-Second Test)
I stripped the design down into a narrative. The rule: No music. Only natural sounds. I want to see if this world makes people feel something without cinematic manipulation.

I built two storyboards:

1. Part 1: A 6-panel story showing the Engineer awakening and seeing the AI for what she truly is—indifferent and vast.

2. Part 2: A 4-panel story showing the psychological weight of receiving a grim prediction.

These are the results.

Narrative: The AI awakens the Engineer. He plugs in, noticing he is a speck before her.

Narrative: The AI chills him to the bone with a coldly calculated prediction: “78.5% probability.”

Step 3: The Execution
We fed both storyboards into Seedance 2.0 to animate them.

This is the end result:

It was a fun experiment. It forced me to define characters and world-build a little bit. Let me know if you are interested in the prompts.

We’ll se how it pans out. ⚡️

Subscribe now

Build worlds. Have fun. Enjoy coffee.

-Erik.

We are spoiled, these days.

Character consistency used to be the biggest headache in AI storytelling. Every new generation felt like a different person wearing the same costume. But not anymore. Consistency is easy now. And we should take advantage of that to make real videos.

I’m writing Kaiju Protocol. Beatrix is my main character. Here’s exactly how I built her style sheet and multiple shot references, so she stays consistent across everything I create.

Subscribe now

The Portrait Grid

Before anything else, I lock in the face. I use a 3x3 grid with nine different portraits showing different expressions and gestures. Hands included where needed.

This gives me a library of emotions to pull from later. Calm. Laughing. Winking. Enraged. Shy. I can reference these during animation and know exactly what the character should look like.

3x3 grid with nine different portraits of character, showing different face gestures and if needed, visible hands.

The Character Sheet

Once the face is locked, I expand to the full sheet.

For Beatrix, I needed:

• Face close-up on top

• Full body front view

• Side view

• Expression sheet with the five key emotions

• Signature blue cyber tattoo on back, arms, hands, belly, not in the face

Feel free to adjust these needs depending on your character. Perhaps there is an artifact (a sword, a medallion or helmet) you need to have defined.

Create a character sheet for an original female warrior named Beatrix Aliger.
Include:
- Face close up on top
- Full body front view
- Side view

- Expression sheet with: calm, laughing, winking, enraged, shy
- Signature blue cyber tattoo in symmetric detail in back, arms, hands, and belly, specify: not in face
STYLE: deduce from references. Match the exact style and finish.

Art direction:
Cinematic premium character design sheet, dramatic action energy, detailed costume design, dynamic composition, expressive hair movement, stylish and badass. The action pose should feel explosive, cinematic, dramatic motion.

Typography:
Use sci-fi character-sheet style titles and labels, including:
“BEATRIX ALIGER”

One tip: GPT Image 2 is like an eager intern. It will add lore text even if you don’t ask. If you hate that (I do sometimes), be clear: specify exactly what text you want.

In my case, I just wanted to specify: “Blue tattoos should NOT appear in her face.”

Good enough.

The Video

Now we take all of that and push it into motion.

I’m using Seedance 2.0 to bring the character sheet to life and make a cinematic trailer. The goal is to translate the “explosive action energy” from the static sheet into actual movement. Beatrix in action, not just in pose.

Let’s go.

Build worlds. Have fun. Enjoy coffee.

-Erik.

Beatrix is exploring a derelict starship in Chapter 1 of my story. The text is already done, posted, and live.

Great. Now, let’s turn it into a video. 🫠

Subscribe now

There are two stages of my current workflow, made possible thanks to two recent tools:

GPT Image 2 — It’s better than Nano Banana 2 at handling text and understanding context. Ideal for storyboards, with and without text.

Seedance 2.0 — The best model I’ve found for handling multishots. The ideal companion for GPT 2.

So how does this work?

Step 1: Provide references of your character and any persistent elements.

Step 2: Describe the events of each shot and the general style.

Step 3: Generate an initial storyboard grid.

Step 4: Add that storyboard to your reference stack for the video model. Adjust as needed.

Is this good?

The Cons: You may get one or two beats you don’t like. You may need to edit those in Capcut. Generations are longer and use more tokens.

The Pros: Better overall sequences because the tool has more context. I usually get great results within 1-2 generations.

Take a look at the prompts and the results:

Storyboard 1

Create a cinematic storyboard in a 3x3 grid (9 panels). Each panel must look like a frame from a high-budget anime feature film. Strong visual consistency across all panels.

Hard sci-fi interior: military-grade metallic construction, amber emergency lighting, vacuum silence.

Style: hand-painted 2.5D illustration, textured painterly brushwork, vibrating grease-pencil outlines, 3D only as structural base, expressive organic linework with visible brush texture, dramatic chiaroscuro, cold blue for space / amber for interior.

Panel 1 — ESTABLISHING SHOT Deep space. A massive derelict military vessel — kilometer-long, dark grey hull, angled wrong in an asteroid field like something that died mid-motion. Asteroids drift at multiple depths. Cold blue starlight. No motion. Scale makes the ship feel like a dead continent.

Panel 2 — PLASMA ARC (EXTREME CLOSE-UP) Two gloved hands hold a [Reference 2] against a blast door seam. Extreme close-up on the arc: molten alloy spits and dies instantly in vacuum — no trail, no smoke, just light extinguishing. The metal edge glows guilty white. The glove shows wear.

Panel 3 — CHARACTER REVEALED [Reference 1] stands in front of the hatch with an exosuit, magnetic boots locked. Holding the [Reference 2] with both hands. She is in a closed room, the hatch is reinforced metal.

Panel 4 — THE HATCH SURRENDERS [Reference 1] drives her gloved hands into the seam of a blast door. The door shudders open — no sound, just darkness beyond. Her body leans into the effort, then releases. The darkness inside is absolute.

Panel 5 — THE THRESHOLD Frontal shot of [Reference 1] opening the hatch with both hands, POV fro the dark corridor. Her helmet lamp creates a single cone of amber light into pure blackness.

Panel 6 — DRONES LAUNCH Rear shot of [Reference 1] Two compact DV-Mini drones detach from magnetic mounts on her suit’s back and float forward into the dark corridor — their scanning lights sweep in steady arcs across military composite walls.

Panel 7 — INTO THE DARK [Reference 1] pushes off a dark corridor in zero-g — a single controlled motion, boots coming free. Side view, She drifts in and rotates slightly, helmet lamp cutting the dark. The motion is practiced and confident. She knows how to do this.

Panel 8 — THE TRAIL Floor-level shot. The corridor floor in the amber beam of a helmet lamp. A rust-brown stain — blood, dried months ago in vacuum — leads from the left edge of frame along the floor and up the wall at hip height, hand-marks trailing lower, lower, dropping to floor level for the final meters. The end of the trail is just off-frame.

Panel 9 — THE BODY (OVER-SHOULDER) Overshoulder of [Reference 1], crouching. Ahead: a dead man, vacuum frozen, slumped on the floor against a sealed blast door. On his side. One arm outstretched toward the door control panel he never reached. His uniform hangs loose on vacuum-dried bone. His fingertips left marks on the panel housing. He almost made it.

Looking good. Let’s turn it into video.

Seedance 2.0 takes the prompt and the storyboard and really understands the mood.

Storyboard 2

Create a cinematic storyboard in a 3x3 grid (9 panels). Each panel must look like a frame from a high-budget anime feature film. Strong visual consistency across all panels.

Hard sci-fi interior: military-grade metallic construction, amber emergency lighting, vacuum silence.

Style: hand-painted 2.5D illustration, textured painterly brushwork, vibrating grease-pencil outlines, 3D only as structural base, expressive organic linework with visible brush texture, dramatic chiaroscuro, cold blue for space / amber for interior.

Panel 1 — EXAMINING THE BODY Medium shot, low angle. [Reference 1] crouches beside dead body. Helmet lamp from above. Corridor stretches behind her.

Panel 2 — FINDING THE DATA CHIP Thumb brushing frost from a data chip clipped to his chest harness. Other hand on his name tag. Helmet lamp from above. Corridor stretches behind her.

Panel 3 — THE OBSTACLE Extreme close-up, insert shot. The blast door panel fills the frame: BIOMETRIC VOICE AUTHORIZATION REQUIRED. Military-grade interface, slightly worn. No character visible. The panel is the entire image.

Panel 4 — BUILDING THE HACK Extreme close-up, insert shot. [Reference 1] ‘s gloved hands assembling a jury-rigged relay device — electronic components floating and being caught and placed with precision. Amber helmet lamp. Focused, technical, deliberate.

Panel 5 — PARTIAL CREDENTIAL ACCEPTED Close-up on the blast door panel. The relay plugged into a secondary port. Panel flickers. Text resolves: PARTIAL CREDENTIAL ACCEPTED Green replaces amber. The door begins to move.

Panel 6 — VAULT OPENS Over-shoulder, static. [Reference 1] stands in the open doorway. Vault darkness behind her. She doesn’t move yet. One beat of stillness before she steps through. The door he died on is open.

Panel 7 — THE AIR BUBBLE Medium shot, interior vault. [Reference 1] hits the atmosphere activation switch on a wall panel. A hiss begins. Vapor fills the rooms. She watches her visor readouts.

Panel 8 — THE VAULT ground view, dramatic view from the ground, interior vault, 3mx3m room. Full body [Reference 1] stands still while the air fills the vault.

Panel 9 — FIRST BREATH Extreme close-up. [Reference 1] removed her helmet, Eyes closing, face looking up. Real air on skin. Not relief — something quieter than relief, something that took five weeks to earn. Clean grey vault walls behind her.

Now the video.

The model held visual consistency across all 18 panels. Same suit. Same mood. Same amber light spilling into absolute dark.

My workflow, today

GPT Image 2 for the storyboard grid.

One edit pass with Freepik AI Suite for the two frames that drifted.

Seedance 2.0 for the motion.

Not bad.⚡️

Build worlds. Have fun. Enjoy coffee.

-Erik.

There is a new wave of AI-native IP builders.

It’s growing.

Gossip Goblin just raised money. Kavan the Kid sold a five-season series. Fruit Love Island hit 200 million views.

Three creators, three completely different approaches to turning generative tools into actual intellectual property. Not memes. Not one-off viral hits. Universes with characters, seasons, and (increasingly) revenue.

Let’s look at what they’re actually doing, what each approach represents, and, more importantly, what you can learn for your own workflow.

Let’s dive in.

Subscribe now

Gossip Goblin (Zack London):
The Auteur studio model

London quit his tech job, raised funding, and launched Gossip Goblin Inc. as a formal studio. He writes and directs everything human-first, anthropology background showing in the world-building, then uses Midjourney, Kling, and Veo 3 as visualization multipliers. The result is bleak, philosophically dense sci-fi shorts that feel like fragments of a much larger mythology.

What this represents:
AI as a force multiplier for existing creative vision, not a replacement for it. He’s generating thousands of images, curating ruthlessly, and building lore-first. The Patchwright and his bunker-society narratives aren’t “AI slop”, they’re deliberately synthetic cinema with its own visual grammar.

The practical takeaway:
If you’re building any serialized universe, this is your permission to go deep on lore before you go wide on content. London’s 600+ episodes exist because he mapped the mythology first. Start worldbuilding with Claude Cowork, explore your character designs in Nano Banana, then produce.

Kavan the Kid (Kavan Cardoza):
The serialized IP-for-Sale pipeline

Cardoza came from commercial photography and music video direction, already thinking in sequences, shots, and deliverables. He used that discipline to build Chronicles of Bone, a five-season fantasy epic with consistent characters, then sold it to Freepik as a Magnific Original. He also landed Echo Hunter as the first SAG-AFTRA-approved AI film, which is not a small detail if you’re trying to look legitimate to buyers.

What this represents:
The solo creator as a sellable production entity. One person, one desk, tools like Seedance and Nano Banana, output that rivals studio pitches. He co-founded PhantomX.ai specifically so creators keep 100% IP ownership.

The practical takeaway:
Your universe isn’t just a creative project, it’s a pitch deck. Kavan didn’t wait for permission; he built the whole season bible, produced proof-of-concept episodes, and took it to market. If you’ve got a world in your mind, you should be thinking now about who buys serialized IP. Freepik. Tubi. Even tool companies like Magnific are hungry for original content to showcase their tech. Build the pilot. Document the pipeline. Make it easy for them to say yes.

Fruit Love Island (AI Cinema):
The algorithm-native meme factory

Anonymous creator, TikTok account, anthropomorphic fruits in a Love Island parody format. Cherrita, Bananito, Strawberrina. Vertical micro-episodes. Coupling ceremonies. Elimination drama. Hundreds of millions of views in weeks.

What this represents:
Speed works as a creative strategy. This is pure format-native content, built for TikTok’s retention mechanics, not for cinematic legacy. AI handles visuals, voices, scripts. The human input is format selection and trend timing.

The practical takeaway:
There’s a real lesson here about platform-native formatting, even if the execution is questionable. Fruit Love Island understood that Love Island’s structure, confession cams, recouplings, cliffhangers, maps perfectly to short-form vertical video. If you’re trying to build an audience, consider: what is the TikTok-native version of your universe? The format doesn’t have to be the final product, but it can be the funnel.

Fair warning:
Fruit Love Island also drew backlash for closely resembling a human creator’s earlier fruit-character series, plus the usual “AI slop” criticism. The growth was real; the sustainability is not. Use this model for audience acquisition, not for your endgame.

What actually matters if you’re building now

Three creators, three lanes. Here are the lessons I’m taking home as I build my own universe, starting with my story Star Wyrms. 🫠

Consistency is non-negotiable. Kavan spent obsessive time on character stability. London generates thousands of images to find the ones that match his visual grammar. Build a visual bible before you build a story bible.

Hybrid workflows win. Every one of these creators uses AI for scale but humans for taste. London uses voice actors and composers. Cardoza layers human sound design. Your workflow should be generative tools AND your judgment, not generative tools alone.

Own the IP, or you’re building someone else’s asset. Kavan built PhantomX specifically for this. London structured Gossip Goblin Inc. to retain control. If you’re using tools that claim rights to your outputs, or platforms that can change algorithmic priorities tomorrow, you’re not building a universe, you’re renting attention. Read the terms. Structure accordingly.

Original IP is the only exit. Cardoza’s fan films went viral but drew takedown threats. His breakthrough came when he sold original work. Fan content is a portfolio piece; original universes are the business. If you’re currently making Star Wars fan fiction, that’s practice. Your sellable asset is the thing that doesn’t exist yet.

Have fun.

GPT Image 2 doesn’t just generate images. It generates the kind of images that make you rethink your entire pipeline.

Which is exactly what I did for the last days, using a biomechanical dragon and five terrified pilots as my test subjects.

There are a few workflows GPT Image 2 opens for creators with defined stories.

You have a draft? turn it into a specific, multi-panel comic layouts with strict color scripts and mechanical detail requirements.

You have characters? turn them into a fighting videogame scene, with a detailed action script.

This is kind of work where Nano Banana 2 starts to smudge the fine print, and GPT Image 2 delivers with ease.

Let’s dive in.

Subscribe now

🤖 Here are some explorations and the exact prompts. Feel free to use them as starting points and modify for your own stories.

UI for website

Create an animated landing page using these image as a reference for style and color grading. The theme is a scifi movie named Star Wyrms

GPT Image 2:

Seedance 2 Video:

Comic Layout

For this one, I used the first part of the Chapter 3 - Resonance of the story Star Wyrms, where the main character meets the Dragons for the first fime.

Page Layout Overview:

Top Tier: Tight, suffocating horizontal strips (the coffin lid closes, darkness swallows them).

Middle Tier: One full-bleed wide shot (the impossible space breathes open).

Bottom Tier: Escalating reveal—impact, scale, then the eye.

Think of the page as an accordion: squeeze → expand → freeze.

Panel 1

Full-width strip across the top (~1.5” high)

Shot: Extreme close-up, side view of the airlock seal.

Comp: Heavy hydraulic bolts slamming home. Scored military metal. Through a tiny viewport in the door, pure blackness beyond. No characters visible yet—just the finality of the mechanism.

SFX: KRAK-CHUNK (deep, mechanical, irrevocable)

Caption: “The airlock sealed behind them with the finality of a coffin lid.”

Color: Cold institutional backlighting from the corridor they left; the edge of the seal catches a dying amber warning light.

Panel 2

Left 1/3 of the second tier

Shot: Kai POV. Total darkness.

Comp: Blackness. At the very bottom edge, we see the toe of Kai’s boot or his gloved hands to ground it as his perspective. Flickering across the black void: his Humanware HUD—orange geometric targeting reticles, a compass ring, system text—all glitching, strobing, and failing. The orange glow barely illuminates his forearms, showing the black suit segments and those orange joint accents from his design.

HUD (faint, stuttering): LOW LIGHT COMPENSATION… FAILED

Caption: “Kai stepped into darkness so complete his Humanware flickered, trying to compensate.”

Note: The HUD is the only light source here. It should feel fragile and intrusive against the void.

Panel 3

Right 2/3 of the second tier

Shot: Close-up on Brandon, beside/over Kai’s shoulder.

Comp: Brandon’s face lit by the dim orange spill from Kai’s HUD. Eyes wide, mouth slightly open. Behind him, the other three teammates are reduced to silhouettes—shoulders, helmets, one rifle barrel catching a glint. They are present but anonymous. The air looks thick, viscous.

Dialogue (whisper, small font): “Holy shit…”

Caption: “…voicing what everyone felt.”

Atmosphere: No breath fog (the temperature is cycling hot/cold, not just freezing), but the background should shimmer slightly, like heat distortion.

Panel 4

FULL-BLEED SPLASH—middle of the page (~4” high, full width)

Shot: Wide establishing. Low angle, looking up from the team’s level.

Comp: The team stands at the bottom edge, tiny—five dark figures, Kai slightly forward. The ceiling is lost in absolute shadow. Raw stone walls rise like a cathedral nave, threaded with bioluminescent strips that pulse exactly like veins—cyan/teal/blue-green light throbbing in a slow rhythm. Organic conduits (fleshy-mechanical hybrid, almost intestinal) wrap through the stone.

Center: The pillar rises, platforms circling it like vertebrae. The mercury moat catches the bioluminescence, silver ripples reflecting the pulsing light.

Caption: “The lights came up slowly… They revealed the full Aviary, a chamber that shouldn’t exist inside a military installation.”

Color Script: Deep indigo shadows. Cyan/teal bioluminescence. Silver mercury. The team’s black suits swallow light; Kai’s orange accents are the only warm pinpricks in the frame.

Border: None. Let the art bleed to the page edges so the reader feels the space expand.

Panel 5

Left 40% of the bottom tier (~2.5” high)

Shot: Ground level, wide angle. The Landing.

Comp: The dragon’s hind leg/claw slams onto the stone platform at the mercury moat’s edge. Mercury erupts in a violent silver splash. Impact cracks spiderweb the stone. The shockwave throws the team off-balance—Kai stumbles forward, arms out, while the others reel. We see only the dragon’s legs and lower torso here: those heavy white/silver armored plates from your reference, the exposed copper/bronze joint mechanisms, claws the size of human torsos.

SFX: BOOOOM (low, reverberating, stone-cracking)

Caption: “A Dragon landed in front of them.”

Scale trick: A single claw tip is wider than Kai’s shoulders.

Panel 6

Right 60% of the bottom tier—THE MONEY PANEL

Shot: Low angle, looking straight up. Kai in foreground, dragon towering.

Comp: Kai is in the lower left, seen from behind or in profile, head tilted back, frozen. His black suit with orange accents makes him the immediate focal point. One hand is half-raised—not quite a shield, not quite a reach.

The dragon fills the rest of the panel, body coiling up and around the central pillar. Its serpentine form is heavy with muscle/armor. The white-to-silver shifting scales catch the bioluminescent light (matte white where shadowed, gleaming silver at the edges). Wings folded tight against its flanks—show the stretched membrane over the mechanical wing-frame (the orange/copper membranes from your reference). The head is lowered directly toward Kai, jaw slightly open. Teeth glow with internal heat (orange-red, the only warm color on the creature).

Internal Caption (Kai): “Kai’s breath caught.”

Text excerpt (worked into the art or as caption): “Fifty meters from snout to tail-tip… scaled in overlapping plates… claws that could punch through starship hull… head all predatory angles.”

Eye note: Your reference shows blue eyes, but the text specifies gold. I recommend the eyes shift from dormant blue (Panel 5, if visible) to active gold here—like a targeting system locking on.

Panel 7

Full-width strip across the very bottom (~1.5” high)

Shot: Extreme close-up. The Dragon’s eye + HUD overlay.

Comp: The eye fills the panel—gold iris, vertical slit pupil. Hyper-reflective, not purely organic nor purely mechanical. In the dark mirror of the pupil, we see a tiny, distorted reflection of Kai. The eye surface has subtle seam-lines and lens-like curvature, hinting at the mecha beneath the biological horror.

Overlaid across the panel in sharp, clinical, glowing orange text (Kai’s Humanware HUD):

┌──────────────────────────┐

│ NEURAL SIGNATURE DETECTED │

│ CLASSIFICATION: UNKNOWN │

│ THREAT ASSESSMENT: EXTREME │

└──────────────────────────┘

Add thin orange HUD brackets and scanning diagnostic lines targeting the eye.

Caption: “And the eyes. Gold. Slit-pupiled. Fixed directly on Kai with an intelligence that made every instinct scream run.”

Effect: The clinical orange text should feel violently out of place against the primal intensity of the eye. System vs. Soul.

GPT Image 2:

Video Game - Character selection

A highly detailed 3D gaming season pass page that shows awesome unlockable characters for a scifi game. Each character a different class.

You can detail as much as you want. For now, I just added the characters from my story Grind Fighter and their names.

Video Game - Fight Scene

Can I direct a fight scene? GPT Image 2 seems to think so.

Use the provided two character images as exact visual references. Preserve identity, face, outfit, proportions, and details perfectly. AAA fighting game scene presented as a Tekken-style gameplay grid storyboard, nine panels showing a continuous fight progression with full game UI visible in each frame.
panel 1: dramatic VS screen — character A vs character B, both facing each other, glowing background, versus text in center;
panel 2: round start — both fighters in idle stance, READY text on screen;
panel 3: FIGHT moment — cinematic spark explosion as fight begins;
panel 4: first attack — character A throws a punch with motion blur;
panel 5: counter attack — character B kicks with impact sparks;
panel 6: combo sequence — multiple hit sparks, motion trails, dynamic camera;
panel 7: low health moment — health bar flashing red, tension moment;
panel 8: finishing move — slow motion heavy hit, cinematic angle;
panel 9: KO screen — opponent falling back, huge KO text, particles flying;

ultra realistic AAA Tekken-style fighting game, cinematic gameplay rendering, highly detailed characters, realistic lighting, in-game fighting camera, slightly angled side view, dynamic zoom during impacts, high-end fighting arena, dark background with neon accents, subtle depth, not distracting.
health_bars: top left and right, detailed, glowing, decreasing across panels;
timer: center top;
player_names: top left and right;
round_indicator: visible;
combo_counter: appears during hits;
hit_effects: sparks, flashes, light streaks;
KO_text: large, bold, center screen in final panel;
READY_FIGHT_text; arcade-style bold typography;

motion blur on attacks, impact sparks, energy trails, camera shake effect, slow motion feeling in finishing hit, screen glow during hits, high contrast, arcade cinematic, vibrant highlights, deep shadows, clean 3x3 grid, perfectly aligned panels, full UI visible, no logos, no watermarks.

GPT Image 2:

And now the video prompt:

Prompt: Use the provided 3x3 grid image as the primary reference for fight choreography, UI layout, camera framing, and progression. Maintain exact character identity and consistency.

Style: ultra realistic AAA fighting game, Tekken-style gameplay, cinematic but in-game fidelity

Camera: side-angle fighting camera, dynamic zoom on hits, subtle shake on impact

Duration: 15 seconds

---

0:00–0:02 — VS SCREEN

Visual:

Close-up cinematic face-off → instant transition to full VS screen

electric energy splits the screen vertically

UI:

- player names top left/right

- health bars fully charged

- timer appears (60)

- subtle glow animation

SFX:

electric crack + bass hit

MUSIC:

cinematic intro stab → rising tension

---

0:02–0:04 — READY → FIGHT

Visual:

fighters snap into idle stance

“READY” appears center → fast transition into “FIGHT” with explosion sparks

UI:

- bold arcade typography

- screen flash + slight camera push

SFX:

whoosh → spark burst

MUSIC:

beat drops (arcade electronic + cinematic hybrid)

---

0:04–0:07 — OPENING ATTACK

Visual:

Character A throws fast straight punch

Character B gets hit slightly → immediate counter kick

Effects:

- sharp motion blur

- bright red/orange hit sparks

- light streak trails

UI:

- health bars start decreasing

- “COUNTER” briefly appears

- small combo counter (1–2 hits)

SFX:

punch hit, armor impact

---

0:07–0:11 — COMBO DOMINANCE

Visual:

Character B launches full combo sequence

mid-air juggle → chained hits → spinning strike

Effects:

- dynamic camera push-in

- micro slow motion on strongest hits

- energy trails following limbs

UI:

- combo counter increases (3 → 5 → 6 hits)

- damage numbers appear

- one health bar drops significantly

SFX:

rapid hit impacts, energy slashes

MUSIC:

tempo increases, aggressive rhythm

---

0:11–0:13 — CRITICAL STATE

Visual:

Character A falls to one knee, staggered

heavy breathing, slight slow motion

UI:

- health bar flashing red

- subtle warning pulse

SFX:

low heartbeat + ambient rumble

MUSIC:

brief silence → tension rise

---

0:13–0:15 — FINISHER + KO

Visual:

Character B delivers final heavy strike (cinematic angle)

impact freezes for split second → opponent launched backward

“KO” appears huge center screen

Effects:

- massive spark explosion

- camera shake

- debris particles + motion blur

UI:

- KO glowing red

- health bar empties completely

SFX:

heavy bass impact + explosion

MUSIC:

final cinematic hit → instant cut

Seedance 2:

🫠 It’s not magic.

GPT Image 2 leaves you feeling you can do much better, if only you add more detail. Maybe. I’ll try.

But for now, the hard-surface detail, the atmospheric perspective, the way my character look in new scenarios. That’s awesome.

The real unlock here isn’t replacing your pipeline. It’s the exploration of new environments and new ideas.

Have fun.

Subscribe now

Knowing you can now build your own fictional universe is liberating.

It’s also a struggle.

What should I do next? What’s the best move now?

I usually go with the most fun move. These days, that's fighting with Seedance 2.0 to make a teaser trailer for my book. ⚡️

Subscribe now

Now that the first draft of Star Wyrms is officially wrapped (Read on Royal Road), I’ve shifted from prose-brain to production-brain. Promoting a book on Royal Road is one thing, but if you want to build a “SoloUniverse” that scales, you need more than a link, you need a hook. I spent the last few days deconstructing what makes a teaser actually work, and then I spent double that time fighting AI to make it happen.

The anatomy of the 30-second Hook

A good teaser isn’t a summary; it’s a vibe check. It isn’t just a bunch of random cool scenes. There should be an intention behind it.

The components of a good Teaser

• The Hook (0–5s): A single striking image or sound that stops the scroll. For me, this is the visual of a “Mecha-Dragon.”

• The Tone/Setting (5–15s): Establish the “vibes.” Is it gritty military or high-flying adventure?

• The Escalation (15–25s): Rapid cuts and rising audio intensity. Show the cost of the conflict.

• The Stinger (25–30s): A final “money shot” followed by the title and a Call to Action (CTA).

With this in mind, I ideated the following idea:

• Main character in a hangar, back view, walking towards its starfighter. Only the sound of his steps. He stops and turns.

• Close up of the dragon metal claws, product shot, black background.

• Star Destroyer looming menacing over a future metropolis/planet

• Close up of the dragon metal muscles and wings, product shot, black background.

• Main character wearing a strange helmet. Nervous sinews twitch around the pilot.

• Close up of the dragon metal head, eye turning on, product shot, black background. The dragon roars

• Fast scene of the dragon attacking the Star Destroyer

Not bad, right?

The Seedance 2.0 workflow

I used Seedance 2.0 for the cinematic movement, but the design work started in the Freepik AI Suite. To keep the ‘Hard Sci-Fi’ look consistent across the teaser, I used Nano Banana 2 to generate the high-fidelity base images for the ‘Product Shots’ before animating them. 🤖

The struggle here is getting the model to understand that these aren’t just lizards and they aren’t just planes. It’s a messy, biomechanical fusion.

For more tips on how to have consistent characters, read here.

Once I had my character, I just needed to define the specific “Product Shots” with Nano Banana 2 🤖.

Macro product shot, centered on the biomechanical dragon [wings]. Lighting: High-contrast rim lighting against a pure obsidian-black background (Void). The [wing] flexes slowly, hydraulic fluid hissing from joint gaps. Grade: Muted industrial, hyper-realistic textures, 8k resolution.

In these cases, “High-contrast rim lighting” does amazing work. It kept it in the “Hard Sci-Fi” realm I needed for Star Wyrms. 🫠

Getting the AI to render organic sinews weaving into a carbon-fiber suit usually results in some Cronenberg-level nightmares. It took about 40 iterations to get a 'visceral but controlled' look for the cockpit sequence that didn't look like a horror movie. 🫠

Extreme frontal close-up inside the pilot’s neural-interface. Metallic, translucent “sinews” and wet biomechanical fibers are weaving themselves around the pilot’s helmet and suit. Lighting: Pulsing bioluminescent blue light synchronized with a heartbeat. The camera follows a fiber as it plugs into a socket on the pilot’s helmet. Visuals: Visceral, claustrophobic, high-detail macro cinematography.

I’m proud of the helmet, btw.

If you’re curious about the specific prompts and references I used, feel free to check out my personal Freepik Space to see the full layout.

Anyway, this is the final teaser:

Let me know what do you think.

The teaser is done, but the real work is the series script. The goal is to move now from “guy with a book” to “guy with a video series” without losing my mind in the process. 🫠

What’s your go-to strategy for maintaining visual consistency when moving from a single “cool” teaser shot to a full-length scene?

Have fun.

Subscribe now

Disclosure: This post may contain affiliate links. If you click on a link and make a purchase, I may earn a small commission at no extra cost to you.

Generative AI video is great, but its consistency is still a massive pain point.

On the flip side, 3D is incredibly consistent, but the learning curve is too steep when you’re trying to move fast.

For a while, it felt like you had to choose between the two. Not anymore. As I’m building out my sci-fi universe, I’m constantly testing new ways to create scenes without losing my mind.

When Freepik announced they integrated Tripo into their arsenal, I had to try it. One of my stories, Star Wyrms (you can read it here), starts in a futuristic hangar where pilots are prepping their Pegasus Starfighters. Getting an AI model to understand the spatial layout of a complex hangar is a very specific kind of solo-creator purgatory. 🫠

Subscribe now

Here are the 3 things I’m doing right now to build 3D environments that actually work:

1. Keep it all in one workflow

I tried Tripo a while ago.The results were good, but the workflow was exhausting. I just can’t handle generating a plate, downloading it, and loading it into another app over and over again. That “download-and-reload” grind is so 2025.

Luckily, platforms are catching on. Freepik adding Tripo natively changes everything. You can generate an image, use inpainting to edit it, set it as your base scene, and start adding elements seamlessly. Using it as part of an integrated workflow is awesome.

2. Build your scene like Lego bricks

When I first started, I tried to prompt a massive, fully stocked hangar with all my Pegasus Starfighters already parked inside, using this image as a reference:

I could practically hear the AI groaning under the weight of the prompt. The result wasn’t usable at all, the ships were a distorted mess. Take a look at those wings:

The fix: Start from scratch and create elements individually. 🛠️

First: I used Nano Banana to edit my base image. I asked it to remove all the ships from the hangar entirely. The moody lighting and atmosphere were perfectly preserved, leaving me with a clean plate to build my scene.

Second: I asked Nano Banana to generate clean orthographic views of a single Pegasus Starfighter. Use this template:

Frontal view of [reference name], white background, reference image for 3d model

Third: I ran those views through Tripo on Freepik to turn the ship into a solid 3D object.

Yeah. A Pegasus Starfighter in actual 3D.

Fourth: Arrange the scene exactly how you need it. Tripo’s interface in Freepik is simple and handles the spatial arrangement easily.

3. Learn to move the camera

This is where the actual magic happens.

Being able to place the camera exactly where you want it, and then prompt the scene based on that specific framing, gives you enormous leverage.

It makes assembling the start and final frames for your video generations incredibly easy:

You are no longer just rolling the dice with your prompts and praying the AI gods understand your vision. You are acting as the director of your own universe.

Now I just need to see if I can apply this exact same workflow to the biomechanical space dragons. I’ll keep you posted.

How would 3D help you in your workflow? Reply and let me know.

Have fun.

Disclosure: This post may contain affiliate links. If you click on a link and make a purchase, I may earn a small commission at no extra cost to you.

While the AI world has been buzzing over the Seedance 2.0 release (and the drama that followed), another heavy hitter has been quietly leveling up. If you’ve been looking the other way, you might have missed that Kling 3.0 Omni just dropped, and it’s honestly on a bit of a heater.

Seedance 2.0 definitely pushed the ceiling for motion realism, but Kling has been delivering feature after feature with impressive consistency. I’m still waiting to get my hands on Seedance, but in the meantime, I wanted to see if Kling 3.0 could handle the “big three” on my wishlist:

• Native audio (Lip Sync included)

• Multi-shot (Stringing different Points of View in one go)

• Dual-mode (Using both references and specific start/last frames)

I’ve been working on a trailer for my recent story, Star Wyrms (you can catch the latest chapters on Royal Road), and it felt like the perfect playground to push these tools.

Let’s go.

Subscribe now

Read more

Quick note: I'm renaming the newsletter to The AI World Builder. Instead of standalone AI video tutorials, everything now serves a bigger mission, building the Neuronomicon, my sci-fi universe. You'll still get the same tool breakdowns and experiments, but now they will be organized around a real, specific project. Here's why.

We are officially entering the era of the One-Person Studio.

The wall between having a story in your head and seeing it on a professional screen has finally collapsed, and for the first time, independent creators don’t just have to write novels, we can build entire visual worlds.

For months, I’ve been hunting for a way to bridge the gap between “cool AI art” and actual, serialized storytelling. It turns out the secret wasn’t just a better prompt; it was a better pipeline. By using a method pioneered by creators like @0xInk_ and @iamneubert, I’ve finally figured out how to translate my characters from 2D concepts into consistent, live-action units.

Currently, I’m developing a Sci-fi Universe. This is the exact “translation” workflow I use to keep my protagonists looking like themselves across every medium.

Subscribe now

The two-step Translation method

Lately, I’ve been using a workflow popularized by awesome creators. It’s a two-phase process that bridges the gap between stylized art and cinematic reality.

Take a look at this example from @0xInk_:

And this one from @iamneubert

Curious? Lets dive in.

The workflow

1. Phase 1: Use a generative model (like Midjourney) to nail the “vibe” and design of the character in an illustrative or anime style. This is where you lock in the silhouette, the gear, and the color palette. This in an important step: Illustrations/anime generations are usually more creative, wild, and willing to explore different aesthetics.

2. Phase 2: Feed that specific design into Nano Banana Pro or Seedream 4.5 to “translate” it into a live-action or high-fidelity 3D version.

It’s basically like hiring a concept artist for the sketch and a world-class cinematographer for the shoot, except you’re doing both from your laptop.

Case Study: Alexandra “Poison” Ivey

In my current sci-fi story Star Wyrms (posted on Royal Road), Alexandra is a pilot for a sentient, biomechanical Dragon unit. I needed her to look “high-fashion military”, sleek, but functional.

The Design (Midjourney): I used a detailed prompt focusing on her “biomechanic suit” and “long red hair in a bun” with the —niji 7 parameter to get that sharp, illustrative edge.

This is the prompt:

A full-body futuristic editorial female character illustration set in a pale neutral studio environment; a sleek, futuristic beautiful woman pilot fair skin with biomechanic suit standing, she wears scifi pilot black suit adjusted to her body, heavy sci-fi belt and bandolier with a side bag and digital controls, sleek robotic military black boots, minimalistic and tight design, long red hair tied in a bun, athletic heroic shape, fit, one robotic long bracelet, sharp and high-fashion. Controlled studio lighting creates crisp rim highlights along the shoulders, jawline, with subtle motion blur suggesting intelligence. Refined yet urgent presence, high-fashion future formal aesthetic in motion, restrained color palette, precise linework, cinematic minimalism. --ar 4:5 --niji 7

And, very important, I used the following images to guide the prompt:

And this is the result:

The Translation (Nano Banana Pro): To get her “real-world” face, I used a simple macro close-up prompt:

Realistic version, live action, photography of woman in the reference image

And if you prefer a 3d version, it’s simpler:

3d version of woman in the reference image

The result? A character that feels like she stepped off a movie set but retains the specific DNA of the original illustration.

Yeah, but about her Dragon?

The same logic applies to the mechs. For the Dragon unit, a silver, armor-plated beast with glowing blue eyes, I went heavy on the “momentum” in Midjourney.

Here is the prompt, using again the same reference images:

Hyper Ultra realistic futuristic towering robotic Dragon with silver armor-plated, long metal bulky long wings, four legs with bulky claws, glowing blue eyes, flying on the void of deep space captured mid-run at extreme speed, body leaning forward with explosive momentum, the robot’s body is adorned with gold bolts and neon lights, contrasting against the metal frame. Futuristic, sleek design, long, powerful tail extends behind the robot, swishing through the air with sharp, smooth segments, The mech’s eyes glow intensely. --chaos 20 --ar 16:9 --sref 1549001673 --stylize 500 --weird 10 --niji 7

Great! Now we turn it into a 3d version:

3d version of dragon in the reference image

Nano Banana Pro was enough to give the metallic plates the weight and texture they needed to feel “real.”

Pro tip: You way want your own reference images. Pinterest is an amazing place to find art you can use as reference. Find the ones you enjoy, and add them to Midjourney as reference.

Nano Banana Pro vs Seedream 4.5

Is Nano Banana Pro the only choice for this workflow? No.

Seedream 4.5 works great too. But it has a specific style. Here’s an example.

Pseudo realism, Full body, Photography of anime female cyber soldier standing with giant cyber gun, latex battle armor, angle view, relaxed mood, abs, magnetic boots, heels, robotic implants, circuits in neck, mood of cool self-containment --chaos 10 --ar 3:4 --stylize 1000 --niji 7

Nano Banana will follow the instruction and will transform the image into a live version, even if it needs to change the lightning of the face. Seedream will try to preserve that detail and the face structure, even if the result is not as realistic.

Pick the one that works best for you.

A quick reality check

Using these tools doesn’t mean the story writes itself (I wish). It just means the “visual tax” of being a creator is lower than ever. You don’t need a $100M studio; you just need a good prompt and a bit of patience.

Enjoy coffee, touch grass, and have fun.

See you soon.

Well, that escalated quickly...

It started in China. A slow drip of awesome AI videos, nothing we haven’t seen before, right? But then the floodgates opened.

We’re talking about Seedance 2.0, the new video model from ByteDance (the folks behind TikTok). It’s currently live in China via their Jianying app, and let’s just say it’s making a lot of noise.

One creator spent 8 hours and 400 yuan (about $55) to churn out a 10-minute prequel called “Hey Mythos,” and the quality is... well, it’s scary.

If you didn’t know it was AI, you’d probably just think it was a high-budget indie film.

Subscribe now

The western “release” (and the panic)

The big news for us? ByteDance says Seedance 2.0 is coming to the global version of CapCut very soon. Some rumors point toward a wider release around February 24th.

But while tech fans are refreshing their apps, Hollywood is reaching for the fire extinguisher. Because there are basically no “guardrails” yet, people have already used it to create:

Tom Cruise taking on Brad Pitt. A rooftop brawl that looks so authentic it prompted Deadpool screenwriter Rhett Reese to post: "I hate to say it. It’s likely over for us."

Why? Well…

Deadpool vs Green Lantern. A cinematic crossover we never thought we'd see, executed with terrifyingly smooth physics.

And An alternate final of Stranger Things. A viral fan-made ending that has Netflix fans doing double-takes.

Yes. All of them flagrant violations of copyright. All of them amazing technically.

“Smash-and-Grab”

The reaction from the industry has been swift and loud. Disney has already fired off a cease-and-desist letter, calling the tool a “virtual smash-and-grab” of their intellectual property, specifically citing the unauthorized use of characters like Baby Yoda and Spider-Man.

The Motion Picture Association (MPA) and SAG-AFTRA have also joined the fray, condemning the “massive scale” of infringement. It’s a bit of a messy situation, especially since Disney just signed a cozy licensing deal with OpenAI, just after Sora 2 did exactly this move.

It seems they don’t mind AI, as long as they’re the ones getting the check.

The specs at a glance

If you’re wondering why the pros are sweating, here is what Seedance 2.0 brings to the table:

• Feature: Use text, images, audio, and other videos as references.

• Director Mode: High control over camera angles, lighting, and movement.

• Native Audio: Generates sound and dialogue simultaneously with the video.

• Length: Currently 15-second clips (👀 ), but they can be stitched into long-form content.

Look, I personally think Seedance 2.0 should respect copyright law. We need a system that works for everyone. However, studios are being a bit naive if they think they can fully control their IP in this new world.

The tech is here. It’s out of the bottle. Every creator now has the tools to build their own universe, and the only real limits left are agency and perseverance.

Grab some popcorn. This legal battle between ByteDance and the big studios is going to be a long one, but the tech isn’t going back into the box.

Last August, I wrote about Genie 3 when Google announced it. Back then, it was a research preview, fascinating, but not accessible.

That just changed.

Google rolled out Project Genie to AI Ultra subscribers last week. It’s not cheap ($250/month) and only for the U.S. for now, but for the first time, creators can actually use this technology.

Here’s what filmmakers are creating with Genie 3, and how it’s different from video generators.

Subscribe now

What is Genie 3?

Genie 3 is a world model that helps users create, explore, and remix interactive 3D worlds. Users can sketch environments with text/images, move in them, and remix existing creations.

Instead of creating a 4-second video clip of a forest, it creates a forest you can walk through in real-time.

What creators are doing with it

Let’s take a look at five examples of what creators are generating with it.

@goodside created this world in 34th Street–Penn Station for a discarded pack of cigarettes.

@fofrAI crafted this kitchen world for a fish trying to escape.

@infiniteyay generated this wondrous world without gravity.

@ClaireSilver12 went beyond that, and used the images for a music video.

Finally, @elder_plinius made the Mario Kart Rainbow Road.

For writers and creators creating films and building fictional universes, Genie 3 offers something unprecedented: you can make the locations from your stories actually explorable.

It’s still not clear what this tool can build, what new genres it will empower. It’s up to us creators to help define that.

Have fun creating. Touch grass. Enjoy coffee.

Cheers.

"Go pick up a pencil." It’s the internet’s favorite comeback for anyone using AI. Usually, it’s meant as a shut-down.

But with Krea’s new Realtime Edit engine, picking up a pencil (or a mouse, or a stylus) is no longer a chore, it’s a superpower. In five minutes, I went from 'stick-figure scribbler' to 'cyberpunk concept artist.'

The most hated meme in art just became the best way to prompt.

Krea Realtime Edit isn’t just a faster generator; it’s a canvas-based interface that allows for instant iteration. By combining text prompts, rough sketches, and image-to-image influences, it shifts the creative process from “batch processing” to “live exploration.”

The power of the Real-Time canvas

At its core, Krea allows you to manipulate visual content instantly. The AI responds as fast as you move your mouse. This is a… dare I say it… game-changer for creators who need to prototype concepts in seconds rather than hours.

To put this to the test, I spent just five minutes exploring different styles and workflows. Here is what Krea Realtime Edit can do:

Subscribe now

1. From 2D concept to 3D render

In this first test, I uploaded a stylized 2D illustration (originally generated in Midjourney) of a cybernetic woman with oversized mechanical fists.

The Workflow: I simply typed “3D art of woman” into the prompt bar.

The Result: Krea instantly interpreted the complex 2D shapes and translated them into a high-fidelity 3D model aesthetic. Notice the lighting on the mechanical gauntlets and the texture of the fabric; the AI maintained the character’s silhouette and pose perfectly while completely overhauling the material properties.

2. From stick figure to masterpiece

This is perhaps the most impressive feature for non-artists. I drew a very basic, beige-colored silhouette of a person waving.

The Prompt: A full-body futuristic editorial character illustration set in a pale neutral studio environment, modern Anime art, Young cyberpunk man, looking at his robotic hand, dressed with a red tshirt, black sport pants, military boots, and black emo messy hair, white background, Controlled studio lighting creates soft rim highlights along her cheekbones, jawline, hair, and the spectral forms, with minimal motion suggesting quiet inevitability rather than aggression. Refined yet enigmatic presence, reimagined through high-fashion futurism, restrained jewel-toned palette accented with soft highlights, precise linework, cinematic minimalism.

The Result: Despite my drawing being little more than a “blob,” Krea used the sketch as a positional guide. It understood that the “arm” I drew was where the robotic hand should be. The result is a clean, professional anime-style character that perfectly matches the complex lighting and fashion cues requested in the prompt.

3. Real-Time graphic design

Logos often require “flavor” that is hard to get right in traditional software without hours of layer masking.

The Prompt: Neuronomicon Logo with fire

The Analysis: I started with a flat, white-on-black text logo. In real-time, as I added the prompt, Krea wrapped the typography in realistic volumetric flames. It didn’t just place fire behind the text; it integrated the glow and the metallic reflections onto the letters themselves, turning a flat asset into a cinematic title card instantly.

All this is cute and everything. But can you create a specific scene for your videos?

Dumdumdumdum…

4. Scene composition and positional control

Finally, I tested how Krea handles complex environments by combining a Midjourney “headshot” with a rough structural sketch.

The Workflow: I placed a generated face in the corner and used “blobs” to define a body sitting on a bench with one leg up. It was a very iterative (and easy) process.

The Prompt: 3d art, woman sitting in the bench of a rusty bar, one foot on top of the table, the other foot on the floor, wearing military brown boots, drinking a beer, night time, bad lights, arm on the back of the bench.

The Analysis: This shows Krea’s ability to “fill in the gaps.” It took the character’s face and projected it onto the body I sketched. It understood the physics of the leg resting on the table and translated my dark scribbles into a moody, atmospheric bar scene with realistic lighting and depth.

Why this matters

Krea Realtime Edit is easy and fast. And it is lowering the barrier to entry for creators who want to experiment without the “prompt-and-wait” fatigue.

It isn’t just about the final image; it’s about the flow state, the ability to see your ideas evolve the moment you think of them.

The landscape of AI video keeps shifting from “silent films” to full-scale cinematic production.

With the release of Pixverse 5.6, the startup introduces a feature that has long been a holy grail for creators: the ability to generate synchronized sound and realistic lip-sync directly from the initial text prompt.

With this move, Pixverse is positioning itself as a realistic competitor to heavyweights like Google’s Veo 3.1 and OpenAI’s Sora 2.

Subscribe now

Great. But is it good?

Glad you asked. To see if it lives up to the hype, I put it to a quick test.

For the challenge, I’m using Midjourney-generated character references. I’m currently in the mood for Saiyan culture, so bear with me.

Here is a breakdown of two distinct scenarios testing emotional range, camera movement, and environmental sound.

Test 1: Dialogue and 360 View

In this first test, I wanted to see how the model handled a “cool” character archetype contrasted with a chaotic environment.

Midjourney Prompt:

Photography, Female Saiyan with Saiyan hairstyle, character concept, cinematic, --chaos 10 --ar 3:4 --exp 10 --stylize 1000

Using this image as a reference, I then used this prompt:

The Pixverse Prompt: Static camera, woman with saiyan hairstyle says with ironic calm “Things are going to get weird. Be ready”. The camera looks away 360, to the scene around her, there are people running scared screaming “Fire!”, cars and buildings on fire, skyscrapers collapsing.

This is the result:

The result: The “ironic calm” requested in the prompt translated well into the vocal delivery, it wasn’t just a generic female voice, but one that carried the weight of the character’s personality. The lip-syncing was tight, matching the cadence of the warning perfectly.

What stood out most was the environmental transition. As the camera panned away from the character, the soundscape shifted from the focused dialogue to the diegetic noise of the city. The screams of “Fire!” are not clear, but there is a low-frequency rumble of collapsing buildings generated in sync with the visual destruction, creating a cool cinematic effect.

Let’s try another one.

Test 2: Dialogue and Zoom out

For the second test, I pushed the model’s ability to handle complex physics, specifically flight, combined with a dramatic camera move and some dramatic emotion.

The Midjourney Prompt:

Photography, Female Saiyan with Saiyan hairstyle, character concept, focused, dramatic background, cinematic, --chaos 10 --ar 16:9 --exp 10 --stylize 1000

The Pixverse Prompt: Slow zoom out while a saiyan woman says with controlled fury “Show me what you can do! I’m ready”. She has Saiyan hairstyle. The camera zooms out to reveal her full body, standing suspended in mid-air, wind moving her clothes, arms to her side, fists ready, feet pointed to the ground, below her cars and buildings are on fire.

The Result: As the camera zooms out, the audio maintains the character’s voice as the primary focus, while the roar of the fires below begins to swell. The “controlled fury” was reflected in a slightly raspy, intense vocal tone.

Visually, the model managed the “suspended in mid-air” physics remarkably well. Often, AI video struggles with characters floating without looking like they are simply “pasted” on a background, but Pixverse 5.6 does a good job integrating the movement of the wind in her clothes with the light from the fires below. The sync between the dialogue and the widening field of view felt intentional. On the bad side, there is a cartoonish look at the end that didn’t go away even as I upscaled.

What do you think?

Pixverse 5.6’s native audio feature is a solid one, and should be a good reason for creators to give this tool a try.

The competition is heating up, and for creators, this means the tools are finally catching up to our imaginations. One by one, models are adding native audio to their tooling. It’s becoming a requirement.

That’s all for today. Have fun, touch grass, and enjoy coffee.

Cheers.

Subscribe now

Freepik is quietly solving one of the most frustrating hurdles for AI creators: consistent, repeatable lip-sync.

While leading models like Veo 3.1 and Seedance 1.0 are impressive for their “all-in-one” capabilities, allowing you to prompt video, sound, and dialogue simultaneously, they have a major flaw for serialized storytelling:

They are great for one-offs, but if you need ten shots of the same character, the voice will vary wildly between generations.

You have control over the lines, but zero control over the identity of the speaker.

Freepik has a new workflow, but it comes with trade-offs.

Let’s dive in.

Subscribe now

A new approach

Freepik is taking a different path. Instead of rolling the dice on a new voice every time, it allows you to pair a specific, established voice with your video prompt. While you can’t upload your own audio files yet, this “voice-first” workflow ensures the persona remains stable.

Check this example:

The Stress Test

To see where the limits lie, I ran three different prompts using the same character in different positions and the same selected voice.

The goal: Keep the voice similar AND have a good lip sync. Let’s test.

Initial Image:

Prompt:

Panoramic shot of a robotic samurai school woman. Fast zoom in to the face of the woman, while she says: “I have a bad feeling about this.” Add soft braam for a sense of dread and danger.

The Result:

Initial Image:

Prompt:

Robotic samurai school woman lands on the ground. Impact blast around her. Fast zoom in to the face of the woman, while she says: “I have a bad feeling about this.” Add soft braam for a sense of dread and danger.

The Result:

Initial Image:

Prompt:

Dynamic movement of Robotic samurai school woman in combat holding a sword. Fast zoom in to the face of the woman, while she says: “I have a bad feeling about this.” Add soft braam for a sense of dread and danger.

The Result:

The takeaway

1. Voice Stability: The tool works exactly as intended. The vocal tone and inflection remained almost identical across all three shots. For creators making series or shorts, this is a massive win.

2. The Lip Sync: It’s very good. The easy workflow delivered great results.

3. Another observation: While the voice stayed the same, the character’s appearance drifted, mainly because of the nature of the shots: The panoramic lacked definition, the bird’s view didn’t have a clear view of the face. In every shot, she looked like a slightly different person. A way to add a character reference would be gold.

This is a simple but good step towards solving the the “audio identity” problem.

The holy grail of video for 2026 will be the moment we can combine these stable voices with consistent character references in a single AI tool.

We aren’t there yet, but the gap is closing.

Cheers.

Subscribe now

Disclosure: This post may contain affiliate links. If you click on a link and make a purchase, I may earn a small commission at no extra cost to you.

We start this 2026 with a bang: Kling Motion Control helps creators master movement.

The core of Kling 2.6 Motion Control is simple: it uses a Video Reference to drive an Image Reference. Think of it as digital puppetry. Instead of hoping the AI understands the nuance of a “cool dance” or a “subtle shrug,” you simply show it.

It’s basically like teaching a toddler to dance. You have to show them exactly how it’s done, otherwise, they’re just going to stare at you and eventually fall over for no reason (worked with my kids).

Subscribe now

So. Breathe and relax.

How to use Kling Motion Control

This is the basic UI interface. Notice how it includes a Library of templates.

1. Select Your “Actor” (Motion Reference): Upload a video of the action you want to mimic. You can use your own footage (perfect for specific gestures) or choose from Kling’s Motion Library.

2. Select Your “Costume” (Image Reference): Upload the character image. Pro Tip: Make sure your character’s proportions (full-body vs. half-body) match the reference video.

3. Choose Your Orientation:

   • Character Orientation Matches Image: Better if you want to maintain a specific camera angle from your original image.

   • Character Orientation Matches Video: Best for following the action exactly.

4. Finalize the Scene: Enter a text prompt to describe the background, lighting and any other element you may want to add. Kling handles the movement; your prompt handles the “vibe.”

3 tips for clean results

While this tool is a massive step forward, it isn’t magic. To get professional-grade results without the weird results, follow these rules:

• Spatial Awareness: If your reference video has a character doing a backflip, don’t use a tight headshot as your image reference. The AI needs “room” in the frame to render the movement.

• Speed Control: Stick to references with moderate speed. Extremely fast or blurry movement in the reference video often leads to “melting” pixels in the output.

• The “Half-Body” Rule: If your motion reference is a person from the waist up, your character image should be too. Mixing a full-body dance with a close-up portrait is a recipe for a very confused AI.

Watch how they are doing it

Creators are already pushing this to the limit.

Check out @lucatac0, who used their own physical acting to drive a Harley Quinn animation.

@Preda2005 showcasing the dance moves from the Library onto a cute baby character.

Finally, @ozan_sihay tested his own gestures on different characters.

Stop settling for random movement. It’s time to start controlling it.

Have fun.

AI workflows have simplified dramatically with the power of Nano Banana Pro. These tools now offer creators a high degree of character consistency, even when dealing with highly complicated elements like intricate armor or specific lighting setups.

The “holy grail” of AI video is keeping your character looking like your character from shot to shot. Take a look at the example below. This workflow understands the complex character (spiky suit, specific facial features) and explores perfectly across nine different cinematic angles.

Subscribe now

So, how do we use it?

I’m going to give you two highly detailed prompts you can use to generate two different style references.

1. The Contact Sheet: A prompt for multiple views of the scene to establish angles.

2. The Director’s Cut: A prompt that pushes for different poses, acting as a basic storyboard tool.

These two prompts are enough to generate a complete scene with a highly consistent character.

Let’s go.

Read more

For a minute, it looked like OpenAI had locked up the crown. Sora 2’s launch in October set a new standard for realism and synchronized audio, seemingly leaving competitors in the dust.

Then, Google threw its counter-punch.

Meet Veo 3.1, the update that just dropped and aims to snatch the crown right back. I’ve been digging into the new model, and this isn’t just a minor update, it’s a direct challenge to Sora’s dominance.

The two top dogs in AI video are now Sora 2 and Veo 3.1. So, how do they really stack up in a head-to-head test? Let’s break down the new features.

Subscribe now

Meet the contenders

• Veo 3.1 is Google’s direct upgrade to Veo 3, built to enhance creative control, realism, and (most importantly) audio-visual sync.

• Sora 2 is OpenAI’s second-gen model, prized for its realism, physics, and ability to generate video with synchronized audio from text, images, or even other video prompts.

Head to Head

The baseline for both models is already incredibly high.

Both Veo 3.1 and Sora 2 can handle Multi-Shot Generation (letting you specify cuts and new shots at timestamps within a single prompt) and Synchronized Audio (generating sound effects, music, and dialogue that matches the action). Seedance 1.0 also competes on multi-shot, but only Veo and Sora are mastering synced dialogue.

This is where things get interesting. Veo 3.1 introduces three new features that are designed to give creators the granular control we’ve been asking for.

1. The Game-Changer: Ingredients to Video

This is Veo’s biggest new weapon. You can now provide multiple reference images, characters, specific items, outfits, props, and Veo 3.1 will integrate them all into a single, cohesive, fully-formed scene with sound.

This allows for unprecedented style and character consistency. You’re no longer just rolling the dice on the model’s interpretation; you’re directing it.

• Sora 2: Since it has a focus on Social media, it has a different spin of this feature: You can upload a video and prompt changes to it.

• Other Models: Kling 1.6 has a similar “ingredient” or “material” feature, but Veo’s integration with synchronized audio puts it in a class of its own.

2. Narrative Control: First and Last Frame

This is a powerful tool for narrative structure. You can give Veo 3.1 the first frame and the last frame of your desired shot, and the model will generate the entire video in between, creating a seamless and often epic transition. It gives you precise control over your narrative’s start and end points.

• Sora 2: Lacks this specific “A-to-B” functionality. You can provide an initial image to start a video, but you can’t define the end point in the same generation.

• Other Models: Most video tools can this A-to B image-to-video, but Veo’s ability to lock in both ends and sync dialogue is a significant directorial tool.

3. Workflow Power: The Extend Feature

Veo 3.1 now has a native feature to seamlessly expand generated clips beyond the initial 8 seconds, creating continuous shots of a minute or more. Each new segment is generated based on the final second of the previous clip, ensuring strong consistency for both the background and any characters in the scene.

• Sora 2: The core model doesn’t technically have this as a named feature. However, the platforms it’s integrated with (like Freepik and Higgsfield) almost certainly provide their own extension and outpainting workflows, and it does generate longer videos up to 12 seconds (which is better for Multi Shots). Veo 3.1’s advantage is having this as a native model capability, which often leads to better consistency.

The final verdict: Who’s winning?

Synchronized dialogue is the current ceiling for AI video, and for a brief moment, Sora 2 was the only model to truly nail it.

With Veo 3.1, Google has not only matched Sora 2 on audio quality but has arguably pulled ahead in what matters most to creators: control. The Ingredients and First/Last Frame features are practical, powerful tools for directors looking to execute a specific vision rather than just discover a happy accident.

Sora 2 is a great model, but Google’s focus on a professional workflow makes Veo 3.1 the new model to beat.

The race is on, and it’s getting faster.

Subscribe now

I tried every trick I know.

For an hour, I hammered away at my keyboard, tweaking prompts with every synonym and sentence structure I could imagine. But no matter what I wrote, “a woman throws a 1-2 punch combination,” “a woman punches a hulking man,” “right cross, left hook”, I simply could not make the woman in my AI-generated video land a knockout blow.

The scene was set, but the action was stuck.

My character refused to follow my direction, leaving her perpetually on the verge of a fight she couldn’t start. It’s a common frustration in AI video generation: getting the look is one thing, but controlling the action is another beast entirely.

Just as I was about to give up, I remembered a specific feature in LumaLabs’ Ray 3: Annotate.

Instead of just describing the action, you can literally draw arrows to guide the motion. Could this be the solution? I decided to put it to the test.

Test 1: A simple retreat

First, I started with something basic. I wanted both opponents to simply step back. I drew two arrows pointing away from the center, assigning one to each character.

The Result:

While not the clean, synchronized retreat I had envisioned, the outcome was fascinating. The model interpreted the backward movement within the context of a fight, making the characters circle each other warily, creating a “dancing” or sizing-up effect. It was an impressive, context-aware interpretation. A solid start.

Test 2: The dodge that became a parry

For my next test, I aimed for more complexity: the man strikes first, and the woman dodges his attack. I drew an arrow for the man’s arm moving forward and a corresponding arrow for the woman’s body moving away.

The Result:

This is where things got interesting. The model followed my first instruction, the man swung his arm. But instead of dodging, the woman parried the blow, blocking his arm with her own. The AI seemed to prioritize a more realistic fight choreography over my specific “dodge” command. The fight continued, but my exact instructions weren’t followed. Close, but not quite there.

The Final Test: The 1-2 Knockout punch

This was the moment of truth. I wanted to see if I could finally create the move that started this whole experiment. My goal: the woman strikes first with her right arm, follows up immediately with her left, and knocks the man down. I drew a precise sequence of arrows mapping out the entire combination.

The Result:

WoooHoo! The animation played out exactly as I had hoped. The woman delivered a swift right, followed by a powerful left hook that sent the hulking man reeling. It was the dynamic, controlled sequence I had been trying—and failing—to achieve with text prompts alone.

Subscribe now

This is the final result.

Here is the final, successful result:

This quick dive into Ray 3’s movement control shows the incredible potential of moving beyond text-only prompts. While it’s not a perfect system, the ability to directly guide the action provides a new layer of creative control for getting the exact shot you need.

For anyone struggling to direct their digital actors in specific action, this tool might just be the knockout punch you’re looking for.

Staring at a blank prompt box, feeling the pressure to be a cinematographer, a lighting director, and a screenwriter all at once? It’s the single biggest barrier for creators using AI tools.

The secret isn’t knowing every technical term. It’s about having a solid structure.

To help you skip the frustration, I’ve deconstructed four production-ready prompts for different use cases. Don’t just read them, steal them. Break them down, swap out the details, and make them your own. This is your cheatsheet.

Subscribe now

Template 1: The MAPPA-Style anime fight scene

This prompt is for creating dynamic, high-impact anime action sequences, focusing on style and energy over complex narrative.

Anime Fight

Format & Style: Cel-shaded anime action, bold linework, high-contrast light, and motion streaks.

Style: 2D–3D hybrid animation in the style of MAPPA and Ufotable.

Tone: tense, heroic, and fast. A female scavenger with metal gloves fights a cybernetic warrior, a duel of pride and skill.

Camera: Dynamic anime camera, handheld feel with parallax.

Notes: Even a great prompt can result in AI quirks (like our character fighting with an arm behind their back). The key is iteration. If one element fails, lock in what works and change the variable that didn’t.

Variants to Explore:

• Change the Style: Swap MAPPA and Ufotable for Studio Ghibli, Hayao Miyazaki, or go full Pixar. Change the tone to get a completely different feel.

• Change the Conflict: Replace the fighters with towering Mechas fighting Kaijus, or Gladiators. Go wild. This shifts the scale from personal duel to epic battle.

• Add Dialogue: Make it your own with dialogue you like: “I will not let you destroy my home!” to give the AI a narrative anchor.

Template 2: The authentic UGC product reaction video

This is designed to generate hyper-realistic, “shot-on-a-phone” content perfect for social media ads or organic-looking marketing.

Fantastic User Generated Content

Format & Style: UGC reaction video, authentic, handheld, shot on front iPhone camera.

Style: Unfiltered realism, slight overexposure, raw and honest. Feels like someone impulsively filming a discovery they’re excited about.

Camera: iPhone 15 Pro front camera in selfie mode. Handheld one-hand shot, slightly shaky with autofocus micro-pulses. Occasional light flare as phone shifts. No stabilization or post edits.

Lens: Native wide lens (~26 mm), full-frame field of view. Deep focus, everything sharp, no artificial blur. Edge distortion from close proximity preserved for realism.

Main Subject(s): One person, late 20s, expressive and energetic. They’re talking fast, gesturing with bottle of neon blue water, making exaggerated facial expressions as they describe how surprisingly good it looks and feels — never taking a sip.

• Wardrobe and Props Casual at-home clothes: oversized white hoodie, messy hair pulled back, natural lighting on face. Props: clear bottle of neon water, slightly frosted from the cold; phone held in other hand; visible countertop behind.

• Location Plain kitchen with daylight spilling through blinds. Foreground: bottle and subject’s hands. Midground: subject’s face centered. Background: out-of-focus fridge, magnets, faint clutter. Camera angle: selfie-style, slight high tilt downward as if chatting with audience.

• Lighting & Palette Pure natural light from side window — unbalanced exposure. Cool daylight tone, soft shadows, slight blue cast from phone auto-balance. Color anchors: skin tone, transparent bottle with neon blue water, white hoodie, gray kitchen, silver fridge. Everything flat and true-to-life — no grade.

• Actions & Camera Beats (0–12 s)
0–4 s — Subject lifts bottle close to camera, eyes wide, talking fast.
> “Guys—look at this water. It’s literally perfect. You can see the light bouncing inside it!” Hand moves bottle to catch glare; tiny rainbow flare appears.
4–8 s — Leans closer, excited whisper.  
> “I swear—it’s so clear it looks fake. Like glass, but colder.” Turns bottle sideways; camera focus flickers between face and label.
8–12 s — Laughs, shakes bottle gently so bubbles rise; gestures toward lens.  
> “No way something this basic feels this fancy. I’m losing it.” Ends with amused grin, half-laugh, shakes head.

• Montage Plan Single unbroken selfie take. Quick handheld adjustments as subject plays with reflection; focus pull moment when bottle crosses lens. No edits, jump cuts, or transitions — full real-time authenticity.

• Dialogue (full) “Guys—look at this water. It’s literally perfect. You can see the light bouncing inside it! I swear—it’s so clear it looks fake. Like glass, but colder. No way something this basic feels this fancy. I’m losing it.” (Laughs, shakes bottle.)

• Sound & Foley Raw phone audio: slight room echo, fridge hum, subtle bottle crackle, fingernail tap on plastic, breathy laugh. Auto gain fluctuates with voice volume; background hiss intact. No music, no noise reduction.

• Finish Completely ungraded iPhone video — flat colors, minor flicker, occasional exposure shifts. No LUT, no color curve, no stabilization. Poster frame: subject mid-gesture, bottle catching daylight spark, mouth open mid-laugh — perfectly imperfect freeze of authenticity.

Notes: I may have gone overboard with the detail and dialogue. It’s too much. But the feel is perfect.

Variants to Explore:

• The Tech Unboxing: Change the main subject to a tech reviewer, late 20s and the props to a sleek new smartphone in minimalist packaging. The dialogue can be about the satisfying click of a button or how premium the screen feels. The tone would shift from excited to appreciative and analytical.

• The “Is It Gross?” Food Challenge: Keep the UGC format but change the props to a jar of bizarrely colored pickles. Change the Actions & Camera Beats to show hesitation, a reluctant sniff, a brave first bite, and an exaggerated reaction (good or bad). The dialogue becomes about texture and unexpected flavors.

• The Luxury Skincare Review: Change the wardrobe to a silk bathrobe and the location to a clean, minimalist bathroom. The props become a glass bottle of face serum. The dialogue should be whispered and focused on sensory details like the cool feel on the skin and subtle, calming scent.

Template 3: The High-Energy cinematic sportswear Ad

This prompt uses a shot-list structure with timestamps, making it perfect for creating fast-paced, rhythmic ads and trailers.

Sports Ad

A 12-second cinematic scifi sportswear ad, high-energy, multi-sport montage in four dynamic cuts.

[0–3s] – CUT 1 / OPEN. Boxing gym: close-up of taped fists with cyber gloves tightening, chalk dust drifting through warm light. Camera pushes in fast as the athlete snaps a jab toward lens. (Audio): sharp thwack, deep bass hit, faint gym echo.

[3–6s] — CUT 2. Indoor running: low tracking shot of running scifi shoes — motion blur, rhythmic flashes of teal light. (Audio): high-tempo beat syncs with pedal cadence, breath layered in.

[6–9s] — CUT 3 • Scifi cyber arena: dusk setting, sodium haze. A warrior trains moves with with a staff, handheld camera. (Audio): sneaker squeak + staff swish + rising synth swell.

[9–12s] — CUT 4 • Close up: frontal shot — scifi warrior face, showing the effort of training. (Audio): high-tempo beat syncs, breath layered in.

Notes: It turned out to be a ultra-modern sport ad, instead of sci-fi. But other than that, defining the timestamps is the most powerful tool to control the video.

Variants to Explore:

• Extreme Sports Edit: Change the settings to a snowy mountain peak, a dirt bike trail, and a crashing ocean wave. The sports become snowboarding, motocross, or surfing. The audio can incorporate environmental sounds like wind whipping and engine roars.

• Focus on Recovery: Shift the tone from high-energy to intense and focused. Replace the action shots with scenes of an athlete in an ice bath, using a massage gun, and doing focused stretching in a quiet studio. The audio becomes slower, focusing on deep breaths and soft, ambient music.

• Athleticism & Lifestyle: Change the setting to an urban cityscape. The actions become yoga on a rooftop at sunrise, sprinting for a subway, and laughing with friends at a post-workout cafe. This broadens the focus from pure performance to an active lifestyle.

Template 4: The kinetic High-Fashion studio shoot

This prompt combines a shot list with detailed descriptions of lighting, gear, and mood to replicate the controlled chaos of a professional fashion shoot.

Fashion Shooting

Format & Style Cinematic fashion editorial reel, fast-paced studio shoot with multi angle cutting and flash-synced transitions.

Style: glossy, modern, Vogue-meets-film energy, controlled chaos of a live shoot captured through kinetic camera language.

Camera: ARRI Alexa Mini LF and handheld BTS cam (Sony FX3) intercut for realism. Mix of dolly tracking, whip-pans, and static tripod bursts. Fast focus pulls between poses, brief shutter smears during transitions.

• Lens Alternating between 35 mm (wide tracking) and 85 mm (macro close-ups). Shallow DOF, rolling focus shifts on every cut for texture. Occasional lens flare from strobes left unmasked for authenticity.

• Main Subject(s) Fashion model — bold, cinematic presence; alternating between slow elegance and high-intensity pose transitions. Facial expressions range from fierce confidence to subtle softness.

• Wardrobe & Props Structured designer look: asymmetric black leather jacket, high-waisted pleated satin pants, gold statement earrings. Props: studio stool, standing mirror, light stand, handheld fan, scattered Polaroids on floor.

• Location Minimal studio with soft-gray cyc wall and two visible strobe umbrellas. Foreground: blurred light reflector. Midground: model posing center. Background: faint silhouettes of crew and gear for editorial realism.

• Lighting & Palette Key: strobe bursts + constant LED fill. Fill: tungsten side spill for warmth. Color anchors: matte black, metallic gold, ivory, deep shadow blue, neutral gray. Alternates between warm and cool tones each cut — high fashion lighting contrast.

• Actions & Camera Beats (0–12 s) 0–2 s — Establishing: Wide dolly-in — model walks toward light flare, hair caught by fan. Flash burst → quick cut. 2–4 s — Pose switch: Medium shot — turns sharply, hands on hips; camera whip-pans, flash triggers mid-spin. Match cut to opposite side angle. 4–6 s — Macro insert: Close-up on eyes under strobe flicker, gold earring swings; shallow focus drift. Flash burst → white cut to next frame. 6–8 s — BTS insert: Behind-the-scenes camera shot — photographer crouching, model mid-turn; overlapping shutter sound. Quick zoom-punch for rhythm. 8–10 s — Power pose: Full-body wide; model leans on stool, jacket open, fan lifts fabric; camera glides in circular arc, slow-motion flash bloom. 10–12 s — Final reveal: Extreme close-up of face; camera rises vertically from collar to eyes, soft exhale; final flash freezes her in white-gold haze.

• Montage Plan Seven rapid cuts tied to strobe bursts and pose changes. Each transition masked by light flash or motion blur. Beat structure: 2–2–2–2–2–2 pacing with subtle slow-mo in last frame. Ends with 0.5-second hold on final flash silhouette.

• Dialogue (if any) Photographer (off-screen, layered across beats): “Yes—hold that! Turn! Flash! Beautiful—reset—chin up, light again!” Model breathes audibly between cues.

• Sound & Foley Layered high-fashion sound design: Flash pops and shutter clicks (sync to cuts). Fan whoosh, fabric ripple, soft heel scuff. Ambient low synth pulse and muted house percussion for pacing (120 BPM). No melody — pure rhythm and texture.

• Finish Fine-grain texture; halation bloom on strobes; rich contrast LUT (gold highlights, lifted blacks, matte whites). Optional subtle split-tone flicker for energy. Poster frame: mid-flash burst, silhouette frozen in white-gold light, camera visible behind — elegant chaos captured in motion.

Notes: This prompt excels at creating a specific, premium aesthetic by layering details about camera types (ARRI Alexa Mini LF), lighting (strobe bursts + constant LED fill), and even off-screen dialogue.

Variants to Explore:

• Vintage Film Noir: Change the style to classic Hollywood glamour, high-contrast black and white. Wardrobe becomes a 1940s trench coat and fedora. The lighting is hard key light, creating long shadows (chiaroscuro). The audio is a smoky jazz saxophone melody and the sound of falling rain outside.

• Gritty Streetwear (On-Location): Move the Location to a graffiti-covered alleyway at night. The Camera is entirely handheld Sony FX3 for a raw, documentary feel. The Lighting is the harsh glare of streetlights and neon signs. The model’s poses are more candid and less structured.

• Ethereal Fantasy: Change the Location to an enchanted forest with glowing mushrooms or a misty, ancient ruin. The Wardrobe becomes flowing silk gowns and fantasy armor. The Lighting is soft, diffused, god rays breaking through trees. The sound design includes magical chimes and ambient natural sounds.

Ooff. Those are a lot of details.

But again, the beauty of Sora 2 and similar tools like Veo 3, is that they will understand what you ask from them, and add the things you don’t define. They can operate with single line prompts, or these gigantic ones.

You define how much control you want. Enjoy.

Subscribe now

I was wrong.

OpenAI has released Sora 2 on the web, and its true power has been revealed. It’s a tool that can be wielded with the precision of a scalpel, not just the randomness of a slot machine. The key is knowing how to talk to it.

Sora 2 can go both ways. You can give a generic prompt and let the AI make all the decisions, or you can specify what you want. Let’s break down both methods.

(I still hate the fact that OpenAI pushed for a social media platform. But I get it. I understand why they are doing it.)

Subscribe now

The two ways to prompt in Sora 2: Chaos vs. Control

After extensive testing, it’s clear Sora 2 operates in two distinct ways. It’s good, like, Veo 3 and Seedance combined good, at both, but only one gives you true creative power.

I tested these prompts with the tool on Higgsfield. It allows for very long prompts, and the UI is great.

1. The “Generic” Prompt: The Slop Machine

This is the initial approach OpenAI showed us. You give Sora 2 a high-level idea, a vibe, and a few keywords, then let the AI fill in the blanks. Don’t get me wrong: It’s fantastic for brainstorming or when you need a cool visual fast, but the results are unpredictable.

Here’s a generic prompt I tested:

A sci-fi dusty movie trailer of future gladiators with robotic arms fighting in cyber arenas. Use the style of initial image, fast-paced editing, multiple cuts, to achieve a futuristic mystical mood. Include background sounds to match the tone, as well as a dramatic narrative voiceover. Close the trailer with a ‘coming soon’ text and include THE movie title “THE GRIND”

The result is a slick, professional-looking trailer. Sora 2 makes all the creative choices on shot selection, pacing, and character design. It’s impressive, but it’s not my vision.

2. The “Specific” Script: For total creative control

This is where Sora 2 becomes a revolutionary tool. By structuring your prompt like a screenplay or a shot list, you can dictate the exact composition, timing, action, sound, and visual effects of your scene.

I used to think prompting a full scene was a gimmick that ceded control to the AI. But with this level of detail, you’re not prompting; you’re directing.

Take a look at this structured script.

Setting: Grey wasteland with ghostly grey mists, no trees, no plants.
Subject: Model — sci-fi female scavenger, white skin, black hair, side-swept bangs, ragged t-shirt, muscled arms.
Lighting: Desaturated moonlight, fog backlit with pale silver tone.
Vibe: sci-fi gladiatorial fight.

[00:00–00:03] Close up of the foot of a robotic warrior crushing a grey rock on the ground. Model can be seen in the distance, facing the newcomer robot.
SFX: deep suspense Braam.
VFX: fog pulses, glimmers with ash.

[00:03–00:05] Fast zoom in through the mist: grey smoke vanishes at frame edges as it glides toward the face of the model, kneeled in the wasteland in combat position. She snaps her head sharply toward the lens, nostrils flaring.
SFX: deep tribal drum, low growl.
VFX: fog pulses, glimmers with ash.

[00:05–00:07] Tracking dolly of the face of Model as Model begins to move, a charge against the robotic rival.
SFX: warped violin string pull, crunch of ground.
VFX: eyeshine flicker, hair snaps with motion blur.

[00:07–00:08] Cut to medium slow-motion side shot: her fist begins to grow robotic glowing circuits, embers crawling up the threads.
SFX: circuits ignition + violin shriek over distant drum hit.

[00:08–00:09] Cut to medium slow-motion close up shot: her body begins to grow robotic glowing circuits in neck and chest, embers crawling up the threads.
SFX: circuits ignition + violin shriek over distant drum hit.

[00:09–00:10] Cut to medium slow-motion frontal shot: the head of the robot has a lone red eye, it starts to glow with intensity.
SFX: glowing red eye + violin shriek over distant drum hit.

[00:10–00:12] Cut to wide slow-motion lateral shot: the silhouettes of Model and robot can be seen in fast motion as they are about to crash, high contrast. Text overlay (glitch reveal, sci-fi style, big bold type): “THE GRIND.”
SFX: heartbeat glitch cut, final drum slam.

The difference is night and day. It’s a lot of detail in the prompt, but Sora 2 was able to integrate a lot of the design requirements, timestamps, camera movements, and sound cues. This isn’t just generating a video; it’s executing a production plan.

How to write your own cinematic prompts

Ready to direct? Here’s how to structure your prompts for maximum control:

1. Set the Scene: Start with global descriptors like Setting, Subject, Lighting, and Vibe. This gives the AI a consistent world to build in.

2. Use Timestamps: Break your scene into timed segments (e.g., [00:00–00:03]). This is the most powerful feature for controlling pacing and cuts.

3. Specify Shots and Action: Be explicit. Use filmmaking language like Close up, Tracking dolly, Fast zoom in, and wide slow-motion lateral shot. Describe the action within that shot precisely.

4. Direct the Audio and Visuals: Use SFX: (sound effects) and VFX: (visual effects) to layer in the details that sell the scene, from a deep suspense Braam to eyeshine flicker.

By adopting this structure, you shift from being a passenger to being the pilot. The tool’s potential is immense, but only if you’re willing to move beyond simple sentences and learn to write for the screen.