MiniMaxH3 Lipsync / TLF2VA

Requires Updated Nova Repos<ul><li><a target="_blank" rel="ugc" href="https://github.com/r-vage/ComfyUI_Eclipse_Nova">Eclipse Nova 1.0.31</a></li><li><a target="_blank" rel="ugc" href="https://github.com/r-vage/ComfyUI_SmartLLM_Nova">SmartLLM Nova 1.1.15</a></li><li><a target="_blank" rel="ugc" href="https://github.com/r-vage/ComfyUI_SmartModelLoader_Nova">Smart Model Loader Nova 1.0.11</a></li><li>added sparse attention but don't use it for lip sync!</li><li>Added a new unload node for the LM Loader. The model stays loaded for reuse during the loop and is unloaded once the loop finishes.</li><li>reuploaded the T2V workflow the duration is controlled in the Smart Folder node now</li></ul>

Files with SHA256: 1ca604345bec267e16175f846756f6f6f6efe81fbb8cb3fb2f85678adf3a1534