How to Restyle Any Winning Ad Into Any Visual Style - [FREE REPO + full guide]

Google quietly shipped a video model that made ad editing one prompt.

It is called Omni Flash, and it does video-to-video. You feed it a finished ad, describe a visual world, and it hands back the same ad rendered in that world.

Same shots. Same camera moves. Same performance and same feeling or vibe that made that ad successful in the first place.

I posted a version of this months ago using Seedance 2.0. That method worked. It also needed a 4-node canvas, a content analyzer pass, a rebuild prompt, and real compute.

Embedded post:

Author: Ori Silver (@OriSilver) Post ID: 2067258603401015606 Source: https://x.com/OriSilver/status/2067258603401015606 Reply to: none

Text:

> http://x.com/i/article/2057021921745649664

This one is 10x faster and it costs a fraction of what the old pipeline cost. You can still use the old workflow it has advanteges but if you want quick cheap and accurat, the free repo here makes it.

When restyling an ad to other formats makes sense?

Testing winning angles just became 10x cheaper, once concept becomes many without a single reshoot.

Fatigue becomes a much longer game, every media buyer has the same graveyard. A winning ad or a proven angle something that printed for 3 weeks until fatigue hits.

The standard response is to go make something new. New concept or new script which you can still do.

That is expensive and it throws away the only thing you actually validated: the angle.

Restyling attacks the problem from the other side. The angle stays. The visual world changes.

Also different audiences have different vibes and styles and it helps you attack many different audiences making different formats.

What creative fatigue actually is

Fatigue is not the audience rejecting your message. It is the audience recognizing your frame.

Same faces, same room, same pacing, scroll past. The idea underneath is still good.

Swap the entire visual language and the algorithm reads it as a new concept, because structurally it is one. The story is proven and the surface is unrecognizable.

The cost math

Producing a new concept from scratch means a script, a shoot or a full generation pipeline, and a coin flip on whether the angle lands.

Restyling means you already know the angle lands. You are only testing surface.

That makes creative testing cheap enough to actually run at volume. 1 winning ad becomes 5 concepts in an afternoon, and every one of them inherits a validated angle.

Any format into any other format

Skill comes in baked with 9 presets (for example)

• Claymation
• Pixar 3D animation
• Pixel art
• Felt puppets
• Voxel

The direction is not fixed. Whatever format you have, you can push it toward whatever format you want to test next.

The one rule that makes it work

Every prompt built by the skill ends with the same verbatim clause:

> keep everything else exactly the same, same shots, same camera moves, same motion

That clause is the whole trick. Without it the model regenerates the ad. With it, the model repaints the ad.

Style details describe materials, surfaces, and rendering. They never introduce a new action, a new character, a new object, or new text. The moment a prompt adds content, it stops being a restyle.

The presets

9 styles ship built in, and the skill builds any style you name on top of the same formula.

• Aardman claymation
• LEGO-style plastic construction bricks
• Felt puppet stop-motion
• 16-bit pixel animation
• Voxel art
• Ghibli-style anime
• Knitted yarn
• X-ray
• Premium Pixar-style 3D

Anything outside that list gets built live through the identical prompt skeleton and shown to you before a single clip renders.

The examples

The repo has 3 side-by-side pairs. Original ad on the left, restyled on the right.

Claymation. Felt puppet. LEGO bricks.

Watch the camera in each pair. It does not move. That is the point.

How it runs

The default path uses the MaxFusion AI MCP. Live capability check, 720p locked, clips returned as inline widgets in chat.

Ads longer than 10 seconds get split into segments, and every segment runs the identical prompt. Consistency comes from prompt identity, not from clever variation.

The flow is gated. Source check, style selection, preservation preferences, prompt approval, delivery choice. Nothing generates before you approve the exact prompt.

If you don't have MaxFusion AI MCP

The pipeline is portable. At the delivery gate you answer no, and the skill hands you the complete run-ready package instead.

You need a Google AI Studio API key for Omni Flash. If your agent can execute against that API, it runs the generation for you.

If not, you get the final prompt per segment plus run instructions. Same gates, same rules, same prompts.

Bottom line

Stop retiring winning ads. Repaint them.

The angle is the asset. The style was always disposable, it just used to be too expensive to throw away.

The skill is free and open source. Full SKILL.md, the 9 presets, the example MP4s, install instructions.

For your agent: https://gist.github.com/holy-templar/4ea9f850351068550c74c164636b3bc6

The full repo: github.com/holy-templar/AD-RESTYLER

I am dropping workflows, repos and skills for free on my telegram channel, usually before it is published here

Join now - https://t.me/primegenai