AI outfit change video from one photo: the silk ribbon fashion film

A popular short-form fashion format goes like this: a model stands in front of a poster-style backdrop, glossy silk ribbons sweep around her, she turns away as the ribbons spin with her, and when she turns back her outfit has changed colour, with the ribbons, the discs and the background changing along with it. It reads like an expensive shoot, but you can make it from one photo.

Why one photo is enough

Image-to-video models animate a still. On their own they drift: the face changes, the outfit melts, the ribbons go wherever they like. The fix is to decide the important frames first and let the video model only fill in the motion between them:

Opening frame: your photo turned into the poster. Same person, same outfit, a clean paper backdrop, silk ribbons and two solid discs in the colours of the outfit.
Middle frame: the same scene with the person turned away from the camera and the ribbons spiralling the other way.
Closing frame: the same composition in the new look: new outfit, new ribbon colours, new disc colour, new backdrop.

An image-editing model (we use Nano Banana 2) makes all three from your photo, so the face, hair and makeup carry over. A video model that takes a start frame and an end frame (Kling 2.5 Turbo) then animates opening → turn and turn → new look as two 5-second clips; because they share the turn frame, they join into one continuous shot. With the turn and the new look pinned, the change looks deliberate, not random.

Prompts that work

Describe the new look concretely. "A dark forest-green tailored blazer over an emerald silk satin blouse" gives a clean result; "something green" does not.
Keep the composition sentence identical across the three frames ("same camera, same framing, same background") so only the outfit and colours move.
Ask for no text in the frames. Type generated inside video tends to wobble; set it afterwards as a separate layer, where it stays sharp and can be in any language.
For the motion, describe one continuous action: she turns away with the ribbons, keeps turning back toward the camera, and the outfit flows into the new look as she comes around. Ask for a locked-off camera and no cuts.

The type layer

The editorial look comes from the type: a big serif word with staggered letters (one rises, one drops below the line), a masthead, a date and two or three short notes in italics. Put the big word over the clothes, not the face, and keep the small type away from busy ribbons. Changing the accent colour at the same moment as the outfit ties the two together.

Do it with an agent

The Silk Ribbon Fashion Film Skill runs the whole thing: you give your agent one half-body, front-facing photo, the new look and the copy.

Make a silk ribbon fashion film from this photo of our model for the autumn drop: she changes from the red leather coat into our emerald suit. Title "Couture", masthead "Maison Kova", date 10/26. 9:16 and 4:5.

It generates the three keyframes and shows them to you first, then the 10-second shot, then renders the type over it in 9:16, 4:5, 1:1 or 16:9. Here is the sample film and its input file. More on agent-made video: Claude Opus 5.5 video generation.

What it costs

The footage is billed as model usage to your own API key: about $0.24 for the three Nano Banana 2 keyframes and about $0.70 for the 10-second Kling 2.5 Turbo video, so under $1 per format at October 2026 prices. We compared newer, pricier video models on the same keyframes and saw no visible difference for this shot. The type layer renders on your own machine, so changing the copy later costs nothing.

Photos and permission

Use a sharp, evenly lit, front-facing half-body photo of one person, with the face and outfit clearly visible. And only use a photo of someone who agreed to appear in your marketing, or stock whose licence covers advertising. The sample on our site uses an AI-generated model, not a real person.

FAQ

Can the colour change go in any direction? Yes. The opening look takes its colours from the photo; the closing look is whatever you describe, from a different colour of the same garment to a completely new outfit.

Why does it sometimes change the face? Each step is generated. Check the three keyframes before the video step and regenerate any that drift; the video follows the keyframes closely.

Can I use it for products? It's built around a person turning. For products, try a product reveal instead.