MiniMax H3 fused turbo: 4-step, 1-minute 5-second video on an RTX 4060 Ti
aziib · reddit · 2026-09-02
A community test of the fused minimax-h3-fused-turbo-int8-convrot model: only 4 sampling steps, generating 0.4MP 5-second video in 1 minute on an RTX 4060 Ti 16GB. Author says the fused torba LoRA makes it better than kijai's fast h3 experimental; uses sage attention, triton, and manualsigmas for speed. Model on Hugging Face, workflow on Civitai.
More from Multimodal
- Endless AI TV channel on one RTX 5090: MiniMax H3 generates faster than it plays — spartong945 · 2026-09-02
- H3-World turns MiniMax H3 into a controllable world simulator with 8k gameplay samples — sachasayan · 2026-09-02
- WeMM-Embedding tops MMEB-v3: 9B scores 59.5, 2B beats every 7B/8B model — tomaarsen · 2026-09-02
- WeMM-Embedding-2B edges out Qwen3-VL-Embedding-8B on MMEB-v2 at quarter size — tomaarsen · 2026-09-02
- Tencent open-sources WeMM-Embedding: unified text/image/video embeddings, Apache 2.0 — tomaarsen · 2026-09-02
- Should style and speed LoRAs go into the latent upscale pass in two-pass workflows? — joseph_jojo_shabadoo · 2026-09-02