Zero Custom Nodes: A ComfyUI Pipeline That Puts a Model Into Any Reference Video's Motion
lmcdesign · reddit · 2026-11-22
The author open-sourced a fashion-video pipeline: give it a reference video of someone moving plus a photo of a model wearing a product, and it renders that model performing the same motion, with original/skeleton/result/upscaled comparisons attached.
- All native ComfyUI nodes — nothing to install besides the models; drag in the JSON and go.
- Stack: SDPose for the skeleton, FLUX.2 klein for the first frame, MiniMax H3 + its Fun ControlNet for video, RelayVSR for upscaling.
- The repo ships 3 importable workflows, links to all models, a step-by-step guide, and a changelog of what didn't work.
- About 27 minutes per 5s clip on a 4090; the author is looking for speedups without quality loss.
- The end goal is folding it into a self-modified open-source tool as a pre-paid-API test-and-approve stage.
More from Multimodal
- AI short film 'The Wow Lunch': Midjourney images animated by Seedance 2.5 via Claude MCP — azed_ai · 2026-11-22
- Wan 2.2 Image-to-Video on 8GB VRAM: 4-Step Clips With Auto Last-Frame Saving — Equivalent-Ad1959 · 2026-11-22
- Open-source Short-Drama Skill helped a creator ship 20+ AI dramas with 50M+ views — vista8 · 2026-11-22
- A burger menu designed as an arcade game screen, prices denominated in HP — gizakdag · 2026-11-22
- AI-generated short film: an AI actress starts questioning her world — THEBLUEHORIZON_AI · 2026-11-22
- pi-H3 distillation test: 4 steps passable, 2 steps break motion and audio — BittiAI · 2026-11-22