Local 4-minute lip-sync video created with MiniMax H3 on RTX 5060Ti
tj-tj-tj-tj · reddit · 2026-08-15
Created a 4-minute-40-second lip-sync video locally (RTX 5060 Ti 16GB / 64GB RAM) using MiniMax H3 and a custom ComfyUI node "TJNODESTUDIOONE".
Implementation:
- Segmentation: Split into 34 sequential 8-second clips using prompts calculated for lyric duration.
- Consistency: Uses a reference image in Reference Mode; otherwise, the last frame of the previous clip serves as the first frame of the next.
- Lip-sync: Synchronized via Audio Lock feature.
- Performance: 12-14 minutes per clip; high local compute load using sege3 + solAttn.
- Accuracy: Current lip-sync accuracy is approximately 85%.
More from Multimodal
- How meloform works: WASM pitch detection plus differentiable circuits learning 1,022 sine coefficients — henloitsjoyce · 2026-10-16
- Documentary filmmaker urges peers to watch 'You Can See Everything' — kyliebytes · 2026-10-16
- Redditor's first Cumfy/Seedance video nails a cinematic one-shot generation — Wave_N_Wind · 2026-10-16
- Kling 4 returns after 8 months: 10 hyper-realistic AI ads for under $5 — CurieuxExplorer · 2026-10-16
- Shanghai AI Lab unveils Intern Lumina U2, a diffusion LLM unifying image, video and 3D tasks — Shanghai-AI-Laboratory · 2026-10-16
- NAVER AI Lab proposes Speculative Forcing to speed up autoregressive video generation — naver-ai · 2026-10-16