https://www.youtube.com/watch?v=I_b2QN-B1W0
Ways to make it faster, or fit with low to no vram:
1.In ComfyUI\comfy\ldm\lightricks\embeddings_connector.py, replace
hidden_states = torch.cat((hidden_states, learnable_registers[hidden_states.shape[1]:].unsqueeze(0).repeat(hidden_states.shape[0], 1, 1)), dim=1)
with
hidden_states = torch.cat((hidden_states, learnable_registers[hidden_states.shape[1]:].unsqueeze(0).repeat(hidden_states.shape[0], 1, 1).to(hidden_states.device)), dim=1)
2. In your run_nvidia_gpu.bat file, add “--reserve-vram 4” or “--novram” to the last line
3. Disable the prompt enhancer
4. IMPORTANT: Use this 4bit Gemma 3 instead of the full version. It will save a ton of memory. https://huggingface.co/unsloth/gemma-3-12b-it-bnb-4bit/tree/main
5. Do not use the native Comfy workflows. They are slower than the ones from ComfyUI-LTXVideo.
