bruxosdovfx - Camera H3 v33
Visual camera editor for MiniMax H3 inside ComfyUI.
Drag the camera, create keyframes, and generate camera-path prompts for H3.
Installation
cd ComfyUI/custom_nodes
git clone https://github.com/NyckM/3d-Camera-control-H3-Minimax.gitRestart ComfyUI and search for:
bruxosdovfx • Camera H3
https://github.com/NyckM/3d-Camera-control-H3-Minimax/tree/main
Models and Lora
https://huggingface.co/NyckM/Meridian_CameraH3_INT8_build_by_BruxosdoVFX/tree/main
FAQ
Comments (15)
For anyone wondering were the git hub link is (since the original is a placeholder), here it is:
you buy ikea, you build the furniture and you wonder what are those additional pieces.
most workflows are like second hand incomplete lego sets.
Cool idea, but the characters freeze in place once the camera moves.
yeap, that this version, for gaussians train
thanks for easily pointing this out
@BruxosDoVFX yep just what i was looking for
@BruxosDoVFX could you recommend me a tool for create the gaussians
@popestmaster I DEV THIS ONE https://github.com/NyckM/BruxosSplat
Where can i download node Sol-Attn (Unversal) used in example workflow? It is not node from: https://github.com/Saganaki22/ComfyUI-sol-attn
From my custom nodes folder I ran
git clone https://github.com/eastmoe/ComfyUI-Sol-Attn.git
then restarted
@nexip48008131
C:\Users\lukas\Documents\ComfyUI\.venv\Scripts\python.exe -m pip install -r C:\Users\lukas\Documents\ComfyUI\custom_nodes\ComfyUI-Sol-Attn\requirements-cuda.txt
...
ERROR: Could not find a version that satisfies the requirement nvidia-cutlass-dsl-libs-base==4.5.3 (from nvidia-cutlass-dsl) (from versions: none)
ERROR: No matching distribution found for nvidia-cutlass-dsl-libs-base==4.5.3
gemini:
This error occurs because nvidia-cutlass-dsl-libs-base does not offer pre-built Windows (win_amd64) packages on standard PyPI. The package includes Linux-native CUDA binaries that pip cannot resolve on a native Windows environment.
Cool, but why can't the character perform other actions after the camera is activated? Is it because the JSON prompt node (which is quite verbose) causes the prompt weight to become too high, leaving too little room for the scene description prompts? I tried placing the scene and character descriptions before your JSON block, but it didn't work—there were only very slight changes (and besides, H3-style rotating camera shots can already be achieved via text descriptions). Please release a V2 version as soon as possible; otherwise, it would be a real shame for such a powerful feature to be limited to just orbital camera effects.
the v2 is ready, and have experimental tools, tha character move is possible, but still in dev, maybe v3
its good but, it could be awesome if ,bro make video playable while controlling camera
V3 gonna have this