Runtime error when an audio reference is provided.
Using the 600 v4 version on the Ref2Video, ema and non ema versions the problem is the same, I can add reference videos no problem, but if I add an audio it does not work, Sampler throws this error: RuntimeError: The size of tensor a (3) must match the size of tensor b (2) at non-singleton dimension 0
Using the 600 v4 version on the Ref2Video, ema and non ema versions the problem is the same, I can add reference videos no problem, but if I add an audio it does not work, Sampler throws this error: RuntimeError: The size of tensor a (3) must match the size of tensor b (2) at non-singleton dimension 0
yeah same error, weird though, for a few runs it worked perfectly fine and then started dropping runtime error. and I am using dedicated sampler and dedicated Lora loader from the dev.
When I use an audio as reference it pulls an error. I found a workflow Minimax H3 - Reference to Video 2 Stage (EN).json and it works pretty well. I changed some nodes to make the turbo lora work properly, but I got the problem fixed.
This is the "tutorial" video. If you have low VRAM don't use the same settings he has, but the workflow is working properly using references audio and audio video references
When I use an audio as reference it pulls an error. I found a workflow Minimax H3 - Reference to Video 2 Stage (EN).json and it works pretty well. I changed some nodes to make the turbo lora work properly, but I got the problem fixed.
This is the "tutorial" video. If you have low VRAM don't use the same settings he has, but the workflow is working properly using references audio and audio video references
looks like Node has been fixed: https://github.com/Larryvrh/ComfyUI-MiniMax-H3-Turbo.git