← sources

MiniMax H3 R2V in ComfyUI keeps the people from the scene reference instead of using my character references

source: reddit (reddit:StableDiffusion)published 2026-10-02 10:39ingested 2026-10-03 02:38

retained text

I’m using **ComfyUI with MiniMax H3 R2V**, and I’m struggling to get it to use the correct people from three reference images:

* **Image 1:** A woman and a man getting into a luxury car.
* **Image 2:** A different man, intended as the male character reference.
* **Image 3:** A different woman, intended as the female character reference.

My goal is to use the scene and composition from Image 1, replace the man with the person from Image 2 and the woman with the person from Image 3, then generate a video of them getting into the car and driving away quickly.

I’ve tried detailed prompts explicitly assigning those roles to each image, but the model keeps generating the people from Image 1 instead of using the identities from Images 2 and 3.

Is this a limitation of H3 R2V, or could it be an issue with my workflow or reference setup? How should I connect and label the images, and phrase the prompt, to separate the scene reference from the character references?

Would it be more reliable to edit the still image first and then animate it? Any working examples or workflow suggestions would be appreciated!

document fetched 2026-10-03 00:32 · canonical https://reddit.com/r/StableDiffusion/comments/1wvqha5/minimax_h3_r2v_in_comfyui_keeps_the_people_from