Image to Video (Mock)

This node produces a native ComfyUI VIDEO output, and it does it without any generative model. The "motion" is a deterministic pan - the selected image gets rolled horizontally a few pixels per frame with a subtle brightness pulse, stitched into a video at your chosen FPS and duration. It's fake motion, and the pack is upfront that it's fake motion, because the purpose isn't to look like a fashion film. It's to make the pipeline behave like it's producing video, so the plumbing - preview, evaluation, MP4 export, manifest - can be built and tested before a real image-to-video provider exists. Milestone 4 (a real provider, either Seedance or FASHN Image to Video per the README) isn't shipped yet.

Mechanically it's a tidy example of ComfyUI's newer video handling. The node builds duration_seconds × fps frames from the single selected image using torch.roll for the horizontal shift, then assembles them via VideoFromComponents with a proper frame rate. What comes out is a first-class VIDEO object - you can preview it in the UI, and downstream nodes (including AFS_ExportRun) treat it as a real video. If you were expecting a generative model here, this is the one honest note to take in: nothing is being generated.

The inputs that matter:

• selected_image and ranking - the winner from Rank Candidates. The ranking isn't used for the pixels; it's how the video record gets its run_id and source_candidate_id provenance, which the manifest needs.
• duration_seconds (default 5, range 1–10) and fps (default 12, range 1–30) - the only knobs that change the output. Twelve FPS at five seconds is a light, exportable MP4; crank duration to 10 and you're just rolling the same pan longer.

Outputs: video (the native VIDEO), video_record (an AFS_VIDEO_RECORD with provider mock, model mock-motion-v1, duration, FPS, frame count - the provenance contract a future real provider will also fill), and video_metadata_json for inspection. Wire video into Evaluate Video (Mock) and Export Run; that's the intended path.

This is part of the zero-cost mock slice, so there's nothing to install beyond the pack itself and no key required - Manager search "AI Fashion Studio", or clone into custom_nodes and restart.

Where people get confused, it's usually the same place: they run it, get a real previewable video, and think they've discovered a free video generator. They haven't - the mock video is a deterministic transformation of the input, not inference. The README is explicit that it's "a deterministic motion preview for orchestration testing." Use it to validate the video chain end to end, build your export and evaluation logic, and then wait for the real provider. When that lands, this node's contract is already in place, which is the entire reason it exists.