Puts your own character into an existing scene frame with one FLUX 3 Image Klein edit pass. The frame is the canvas: the pose, the other people, the camera, the light and the room stay; only the woman is replaced by the character from your reference picture. About 10 seconds per image on an RTX 4090, core ComfyUI nodes only (ComfyUI 0.41+).
Three modes, one link to change: A - face visible (swap); C - oral scene (swap + extra keep rule); B - no face visible (hair only, no face is ever drawn, so she is not turned toward the camera).
Measured on an RTX 4090: 10-13 s per image at 1.0 MP, about 14 GB VRAM peak. Scene preservation (SSIM against the original frame) 0.74-0.88, versus 0.58-0.78 for our Qwen-Image 2.1 recipe on the same frames. Honest trade-off: Qwen takes more face detail from the plate, FLUX 3 keeps the scene more faithfully and is faster.
Included: the workflow (UI + API JSON), the prompt kit (instruction texts, templates, vision questions), README with steps, tuning and troubleshooting.
Requirements (weights not included): flux3_image_klein_sd_int8_convrot.safetensors (diffusion_models), qwen3vl_4b_int8_convrot.safetensors (text_encoders), flux2-vae.safetensors (vae). The weights are under the FLUX Kommunity License - check that it covers your use.
Limits: one woman per frame; in hair-only mode seeds matter, render again if old hair stays. Adults and fictional characters only.
Description
v1.0 - first release: three modes (A face, C oral, B hair only), 12-step distilled int8 FLUX 3 Image Klein, reference-latent edit of scene and plate, output at the scene's size, UI and API files, prompt kit, validate.py; measured on an RTX 4090.