Character Swap Minimax H3 (BFS Collection)
retained text
I’ve released my MiniMax H3 Character Swap LoRA I trained this LoRA to replace the main character in a guide video while trying to preserve the guide’s motion, expressions, camera, and scene. It was trained with guide latents, so the recommended setup is to connect the source video through Add Guide. The ComfyUI Planner workflow is included, but it’s optional. It’s still experimental and far from perfect, it’s a LoRA, not a universal character-swap solution. It works better with people and human-like characters, and results can vary depending on the video, reference image, prompt, and settings. I’ve shared a few examples below so you can see where it works and where it still struggles. The primary LoRA and workflow are available here: [https://huggingface.co/Alissonerdx/Minimax-H3-ComfyUI](https://huggingface.co/Alissonerdx/Minimax-H3-ComfyUI) [https://civitai.red/models/2027766/bfs-best-face-swap?modelVersionId=3401556](https://civitai.red/models/2027766/bfs-best-face-swap?modelVersionId=3401556) Feel free to try it and share your results or feedback. https://reddit.com/link/1x4jh3c/video/ilctuqgwx4vh1/player https://reddit.com/link/1x4jh3c/video/x7wryiexx4vh1/player https://reddit.com/link/1x4jh3c/video/maqraszyx4vh1/player https://reddit.com/link/1x4jh3c/video/wkrdiwe0y4vh1/player https://reddit.com/link/1x4jh3c/video/zwh2gzj1y4vh1/player
document fetched 2026-10-14 00:32 · canonical https://reddit.com/r/StableDiffusion/comments/1x4jh3c/character_swap_minimax_h3_bfs_collection