Find the clothing
The person media is analyzed to identify a shirt, coat, or jacket.
Turn clothing into a moving screen. The effect identifies the shirt or jacket, keeps the person and scene intact, and plays your second video only inside the clothing area.
Paste one public person image or video URL and one public texture-video URL that Replicate can fetch.
The effect uses clothing parsing instead of replacing the whole person, so the face, background, and surrounding frame remain visible.
The person media is analyzed to identify a shirt, coat, or jacket.
The detected clothing becomes a stable area for the visual treatment.
Your texture clip is resized and played only inside the clothing mask.
Edges, opacity, and optional source lighting are combined into an MP4.
Use a clean front-facing person shot and a contrasting shirt or jacket for the clearest result. Complex poses and occlusion can reduce mask accuracy.
Put moving graphics, material studies, or campaign footage inside a look.
Turn a performer’s top into a contained layer of rhythm, texture, or archival footage.
Create a transformation that reads quickly in vertical social video.
Test animated clothing treatments before a larger compositing production.
Hard edges and full-opacity video for a graphic screen-print feel.
Softer edges with original highlights retained over the moving texture.
A slightly expanded mask that makes the inserted video more dominant.
The stable BLXSSXM endpoint keeps the existing garment-video-fill API slug for compatibility while the public effect uses the clearer Clothing Video Fill name.
POST /api/effects/garment-video-fill
{
"personMediaUrl": "https://.../person.mp4",
"textureVideoUrl": "https://.../texture.mp4",
"preserveLighting": true
}It detects a shirt, jacket, or similar upper-body clothing and composites a second video only inside that detected area.
Use a public image or video URL containing one clearly visible person and a public texture-video URL. Direct uploads are available in the BLXSSXM iOS app.
Shirts, jackets, and coats with clear boundaries work best. Low contrast, crossed arms, heavy occlusion, and very loose silhouettes can make the mask less accurate.
Yes. POST to /api/effects/garment-video-fill, then poll the returned prediction id through the matching status route.
No. This is an experimental effect based on semantic clothing parsing. Fast motion, unusual poses, or hidden clothing may produce unstable edges.