# Nuva Lab > Nuva Lab accelerates, customizes, and deploys open-weight video models. It runs MiniMax H3 and FastH3 on dedicated hosts for businesses, including US companies, with the customer's data and one flat hosting cost. ## What Nuva Lab offers - Dedicated deployment: MiniMax H3 or FastH3 on a host that only one customer uses. Illustrative GPU cost is about $0.014 per second of 768p video at full use (8x RTX PRO 6000 at $3 per GPU-hour, service fees excluded). The MiniMax API lists MiniMax-H3 at $0.08 per second at 768P (October 8, 2026). - Customization: post-training and fine-tuning of the model on customer data. - US deployment: the public MiniMax H3 Community License excludes the United States. Nuva Lab holds a MiniMax authorization for H3 deployments that serve US companies. - FastH3-Omni (open weights since October 20, 2026): the open-weight, accelerated MiniMax H3 with full omni reference input. It takes text, up to 9 reference images, up to 3 reference videos and up to 3 audio references (up to 12 files), first and last frame keyframes and clip continuation, and generates 5 to 15 seconds of video at 24 fps with synchronized audio in one pass. FastVideo measured 5.6–11.9× lower end-to-end latency than MiniMax H3 on 8 NVIDIA GB200 GPUs across 27 matched reference workloads. Nuva Lab built it with FastVideo and runs a production version with dedicated deployment, fine-tuning on customer data and the Creative Agent. - Business and creative agents that run video workflows and improve from production feedback. ## Pages - [FastH3-Omni open weights release](https://nuvalab.ai/fasth3/): the October 20, 2026 open release (same page at https://nuvalab.ai/blog/fasth3-omni-open-weights-release/); who built it (Nuva Lab with FastVideo, on the MiniMax H3 base model); samples; how to run it locally; how Nuva Lab runs it in production on a dedicated host with fine-tuning and the Creative Agent; open weights vs production comparison. - [MiniMax H3 API price vs self-hosting cost](https://nuvalab.ai/blog/minimax-h3-api-price-vs-self-hosting/): MiniMax H3 and Seedance 2.0 API price per second, a private dedicated host at about $0.014 per second at full use, the difference between a stateless API and a dedicated host, and a calculator. - [Fine-tune MiniMax H3 on your brand data](https://nuvalab.ai/blog/fine-tune-minimax-h3-lora-brand-data/): LoRA vs post-training vs FastH3-Omni references, and evaluation on your review criteria. - [Agents and the production feedback loop](https://nuvalab.ai/blog/video-agents-production-feedback-loop/): Creative Agent and FDE Agent, and how approved work guides tuning inside one deployment. - [Dedicated MiniMax H3 deployment for enterprise](https://nuvalab.ai/minimax-h3-dedicated-deployment/): private H3 and FastH3 host, full concurrency, customization, agents, data flywheel, MiniMax authorization for US companies. - [MiniMax H3 inference speed](https://nuvalab.ai/blog/minimax-h3-inference-speed-steps-generation-time/): steps by tool, measured generation time by GPU, FastH3 timing, other speed-up methods, with sources. - [Run MiniMax H3 locally](https://nuvalab.ai/blog/run-minimax-h3-locally-gpu-vram-requirements/): model size, VRAM and RAM by GPU, Mac and multi-GPU support, local vs dedicated host. - [Optimized FastH3 serving](https://haoailab.com/blogs/uniserve-fasth3): production serving runtime for FastH3 on NVIDIA Blackwell, with optional NVFP4. ## Attribution and licensing FastVideo is an independent Apache-2.0 open-source project. FastH3 uses MiniMax H3 weights and derivatives governed by the MiniMax H3 Community License, not Apache 2.0. Nuva Lab is the production partner: FastVideo credits it with production-aligned post-training insights from creative video-agent workloads, dataset curation for FastH3-Omni, and production hardware and traffic for the UniServe serving engine. FastVideo owns and maintains its framework and FastH3 artifacts.