October 8, 2026
MiniMax H3 API price vs self-hosting: the real cost per clip
MiniMax H3 and Seedance 2.0 API price per second vs a private dedicated host at about $0.014 per second. Calculate your cost per accepted clip.
Blog
Nuva Lab blog
Notes from running open-weight video models in production: what they cost, how to customize them, and how agents turn a model into a production line.
Posts
Production video, open-weight models and cost
October 8, 2026
Fine-tune MiniMax H3 on your brand: LoRA, post-training and references
How to fine-tune MiniMax H3 on your brand data: LoRA vs full post-training vs FastH3-Omni references, and tests on your own review criteria.
October 8, 2026
From video model to production line: agents and the feedback loop
A deployed video model is a machine. Agents run production tasks on it, and approved and rejected clips guide the next round of tuning inside your deployment.
October 9, 2026
MiniMax H3 inference speed: steps, generation time and how to make it faster
How many steps MiniMax H3 uses, how long a clip takes on H100, B200 and RTX GPUs, and how FastH3 cuts 49 model calls to 8. Published data with sources.
October 9, 2026
Run MiniMax H3 locally: GPU, VRAM and RAM requirements
MiniMax H3 VRAM requirements by GPU: 8 GB with FastH3 Trim, 12 GB with FastH3 FP8, 32 GB on an RTX 5090, 4× 80 GB for full BF16. Time per clip on RTX, DGX Spark and Mac.
October 20, 2026
FastH3-Omni: Your Open Source “Unlimited Seedance” is here
Open weights and training code. Who built it, how to run it locally, and how to run it in production on your data.
Run open-weight video models in production
Tell us about your video volume and workflow.
Get started · 1 min