This ComfyUI tutorial walks you through a focused outfit-swap workflow powered by the GeminiVideoOmni node. You load a person video with LoadVideo (Your Footage) and a single outfit photo with LoadImage (Outfit Reference). The GeminiVideoOmni node takes these inputs plus a plain-language instruction (provided via a PrimitiveNode) to map the outfit from the reference image onto the moving subject in your clip, while preserving identity, motion, camera framing, and the original background. The generated result is written out with SaveVideo.

Technically, the workflow conditions a video-aware generative model on both the source frames and the reference outfit image. The source video guides structure and motion so faces, hands, and environment remain stable, while the reference guides the appearance of clothing across frames for temporal consistency. The MarkdownNote nodes in the graph provide inline guidance; they don’t affect processing. With just a few nodes and no manual masking or rotoscoping, this setup is ideal for fast virtual try-ons, fashion concepting, and styling previews.