Live-Action + Blender + Wan 2.2 Lora Hybrid Test
retained text
I directed this, composited it in Flame, and did all the ComfyUI/Ai Toolkit stuff that fed into those composites. I was teaching myself the ComfyUI side as I went. It's basically a cracked out version of the old MDMZ Wan 2.2 workflow, modified to use a custom LoRA and work on 12-bit Sony Log 3 footage. My buddy designed all the CG animals in Blender, and stills of them were used to train custom LORAs that were then mapped to Sethward's live-action face. SammieRoto was used for the keying and paintouts. I've been lurking here a long time and can't thank you guys enough for the guidance! Happy to answer any questions about how it was done.
document fetched 2026-11-10 00:30 · canonical https://reddit.com/r/StableDiffusion/comments/1xrdsj8/liveaction_blender_wan_22_lora_hybrid_test