Troubleshooting ComfyUI: Why T2V Generation is Slower Than Ref2V
5tephaniehemming · reddit · 2026-08-11
The author noticed that in their local ComfyUI setup, text-to-video (T2V) generation takes twice as long as reference-to-video (Ref2V), occasionally hanging after loading MiniMaxH3AudioVAE.
Running on an RTX 3090 and 32GB RAM, they launch ComfyUI with flags like --windows-standalone-build --reserve-vram 1 --disable-pinned-memory --fast fp16accumulation. They are asking the community for tips to resolve this performance anomaly.
More from Multimodal
- MiniMax H3 Generates Hilarious Pigeon Version of Tommy Wiseau — cocktailpeanut · 2026-08-11
- Meta Returns to Open Weights with Muse Glimmer (30B, Apache 2.0) — Simon Willison · 2026-08-11
- Developer Showcases Scene from Upcoming AI Video Generation Project — lmoroney · 2026-08-11
- Why Haven't Chinese Labs Distilled Suno and Released It as Open Weights? — breath_mirror · 2026-08-11
- Vizard Launches Video Agent: Turn Raw Footage or Ideas into Finished Videos Automatically — LearnWithBishal · 2026-08-11
- Maker Builds Custom Imaging Device for 3D Flower Models — NikoMcCarty · 2026-08-11