How I got an AI video model to change outfits exactly on the fan blade

Description and use

Using Seedance 2.0 Mini’s reference mode, outfit changes were synchronized with fan blade passes, achieving 100 % accuracy for up to three outfits.

Problem

The AI video model could not synchronize outfit changes with fan blade passes when only text prompts were used.

Outcome

By providing a reference video that demonstrates the correct fan rhythm, the AI model could reliably synchronize outfit changes with blade passes for up to three outfits, achieving 100% accuracy in those cases.

Workflow

• Upload one photo per outfit to BlinkPop.
• Compose the scene with an overhead shot, carpet, sofa, and the first outfit image using an image model.
• Animate the stills into a short vertical clip with Seedance 2.0 Mini.
• For timing, create short rhythm clips from previous renders showing correct fan blade passes.
• Send the outfit stills as reference_image_urls and the rhythm clip as reference_video_urls to Seedance 2.0 Mini.
• Include a prompt that copies camera, fan, and timing from the reference video and uses the images for room, person, and outfits.
• Receive the rendered video and verify outfit changes align with blade passes.
• If the job fails, automatically refund credits and allow the job to proceed if the photo check service is down.

Evidence

The first version gave the video model the stills and a prompt like "a ceiling fan turns slowly; each time a blade passes the lens, the outfit changes to the next one."

Source

Seedance 2.0 Mini has a reference mode: besides reference images, you can pass a reference video, and it copies camera, motion and timing from it.

Source

Results: 2 and 3 outfits: 8 out of 8 correct (man, woman and a dog)

Source