Create video with synchronized audio on rented GPUs
Deploy the open MiniMax H3 base checkpoint across four GPUs, generate a 768p video and audio result, and avoid the hosted 2K modules.
Before you begin
- Difficulty
- Advanced
- Software
- SGLang, Hugging Face CLI
- Hardware
- The publisher’s FL2VA SGLang example uses four GPUs. Exact GPU memory capacity and throughput for a rental need validation before booking.
Sources and files
MiniMax H3 checkpoint and 768p instructions SGLang H3 deployment guideChoose a model for this task
The steps below use the recommended model. Alternatives have their own package and command; open their model pages before switching.
The workflow
Select the self-hosted 768p path
Use H3-Base FL2VA for text-to-video with audio, or first/last-frame-to-video with audio. The separate 2K context and regeneration modules are hosted by MiniMax and are outside this self-hosted setup.
Provision a four-GPU Linux machine
The publisher documents SGLang with four GPUs and Ulysses degree 4. Before renting, confirm each GPU and the combined machine support the checkpoint and the current SGLang deployment guide.
Install the serving environment
Install a supported SGLang release and Hugging Face CLI on the machine. Keep the server private or behind your own authentication; do not expose the example port directly to the public internet.
Download only the FL2VA checkpoint
Use the publisher’s scoped download so Ref2VA weights are not fetched for this text-first task.
hf download MiniMaxAI/MiniMax-H3 --include "model_index.json" "FL2VA/*" --local-dir MiniMax-H3Start the local H3 service
Use the publisher’s four-GPU FL2VA serving flags. Read the SGLang deployment guide for current package versions and client requests.
sglang serve --model-path MiniMaxAI/MiniMax-H3 --num-gpus 4 --ulysses-degree 4 --performance-mode speed --host 0.0.0.0 --port 30010 --model-variant fl2vaGenerate and check a 768p example
Run MiniMax’s linked T2VA example request against your SGLang endpoint. A successful result includes a playable video and synchronized audio at the H3-Base output resolution. Do not describe this as the full 2K pipeline.
When it doesn’t go to plan
Confirm all four GPUs are visible to SGLang, the FL2VA files downloaded completely, and the installed SGLang version matches its H3 deployment guide.
Use the publisher’s FL2VA T2VA request script for synchronized audio. The self-hosted base path produces 768p; the 2K modules require separate MiniMax-hosted services.
The model behind this workflow
MiniMax H3Will it run on your machine?
Save your machine to see a personalized rating and its reasoning.
Add my hardwareKeep exploring
Set up a private AI chat assistant
Install a local runtime, run Qwen3.5 9B, confirm responses, and know when to choose the smaller 4B package.
Build a private local coding assistant
Connect an open coding-capable model in Ollama to Cline, run a small repository task, and review the result locally.
Generate images on your own GPU
Use the official FLUX.2 Klein 4B ComfyUI template with exact model files, a first prompt, and an output check.