A fully open-source lip-synced music video: LTX 2.5 + ComfyUI workflow with all settings shared
Important-Apple7422 · reddit · 2026-11-01
A Redditor shared his first music video made 100% with open-source models running locally, with the full workflow published.
- Song: YuE2; singer and start images: Qwen Image 2.1; cinematic shots: Pika H3; prompts drafted with ChatGPT/Claude
- Used Codex to split the audio at natural vocal breaks instead of equal chunks; Mel-Band RoFormer separates vocals, and only vocals condition the video
- Lip sync via LTX 2.5 image-to-video in ComfyUI: Gemma 4 12B text encoder (LTX 2.5 finetune), two-pass sampling (8-step distilled eulerancestral at 448x800, then 2x latent upscaler + 3 steps), 25fps, 14s clips per vocal phrase, tiled VAE decode, locked tripod camera so motion comes from lips and jaw
- Hardware: RTX 4090; fast generation allowed unlimited re-rolls per shot
- He first tried OpenArt but got poor results and burned tokens fast; going local meant no credits or per-generation costs, and the process was simpler than expected
More from Multimodal
- Kimi autonomously used a 4x3090 rig and YuE2 to score its own 45-second launch film — minchoi · 2026-11-01
- Ghibli-style hand-drawn 3D game built in one day: Kimi models assets via Blender Python — minchoi · 2026-11-01
- Kimi 3D Make turns one reference image into a 3D model in ~2.5 minutes — minchoi · 2026-11-01
- A Rolex Day-Date built in pure Three.js code via Kimi — 32 minutes, $6.25, no imported models — minchoi · 2026-11-01
- One prompt to real 3D: Kimi drives Blender, CAD and Three.js in 10 wild community demos — minchoi · 2026-11-01
- 2001 forum post predicted the "AI Movie Player" — 25 years later it's real — cronocr · 2026-11-01