Give it a video and a camera offset — azimuth, elevation — and it generates the same scene from that new viewpoint.
This is the CrossView-Warp idea ported to MiniMax-H3. The model reads two videos: a depth-warp of your clip, which carries the geometry, and the clip itself, which carries identity and appearance. The warp comes from the CrossViewWarp ComfyUI node.
Usage (ComfyUI)
LoRA strength: 0.8 - 1.0
Trigger word: crossview (note: You can also influence the generation by prompting what you want to see in the unseen (magenta) areas.)
Example ComfyUI workflow: crossview-warp-h3.json
The CrossViewWarp node has a walkthrough video that covers the camera controls; everything there applies unchanged:
You can find more details in my HF repo: https://huggingface.co/Cseti/MiniMax-H3_Ref2VA-LoRA-CrossView-Warp_v1
Support
Everything here is open, and the GPUs behind it are rented. If this was useful, please consider supporting my work:
Description
FAQ
Comments (8)
Woohoo! Glad to see you're at it again in H3! Loved your LoRAs for LTX.
Keep up the great work.
It seems there is significant image loss in the reconstruction?
I don't feel so. But it you experience any, probably you can fix it in a 2nd pass.
Check your workflow mate. Things dont' automatically work just becuase you copy and paste prompts and stuff. You might need to adjust weights, try different models, and so on. There are too many variables to just rely on copy pasting someone else's workflow and thinking it'll work the exact same.
magic stuff
This is awesome!!