Skip to content

Deploy MiniMax H3 on RunPod

Use this page to choose a MiniMax H3 template, deploy it on a suitable NVIDIA GPU and open ComfyUI. All four templates generate video with audio; the main choices are the workflow type and whether to include Qwen-VL prompt enhancement.

Quick recommendation

  • If you are unsure which template to use, start with MiniMax H3 FL2VA + Qwen-VL. FL2VA supports text-to-video, image-to-video and first/last-frame-to-video, while Qwen-VL provides optional local prompt enhancement.

  • Select an L40S if you are unsure about which GPU to use.

  • Check the tested hardware profiles.

  • Choose a template.
  • Configure and deploy the pod.
  • Wait until the container log reports that the pod is ready.
  • Open Connect → HTTP Services → ComfyUI (port 8188).

Choose a template

Choose FL2VA for general video generation or Ref2VA when reference media should guide the result.

Workflow Use it for Accepted inputs
FL2VA Text-to-video, image-to-video and first/last-frame-to-video Text with an optional first frame, last frame or both
Ref2VA Reference-guided video generation Text with reference images, video and/or audio

Optional Qwen-VL prompt enhancement

The enhancer expands your input prompt before generation. It is optional and does not replace the MiniMax H3 inference model.

Enhancer How it works Choose it when
Qwen-VL Uses a separate Qwen-VL model with llama.cpp You want local prompt expansion and do not mind loading an additional model

Choose a base template if you write prompts yourself. Qwen-VL is the only prompt-enhancement variant offered by these templates. Your chosen workflow, resolution, duration and GPU have the larger effect on video-generation time.

Deploy a template

Template Recommended use RunPod
FL2VA + Qwen-VL Best starting point for most users Deploy FL2VA + Qwen-VL →
FL2VA base General generation with manually written prompts Deploy FL2VA base →
Ref2VA + Qwen-VL Reference-guided generation with Qwen-VL prompt enhancement Deploy Ref2VA + Qwen-VL →
Ref2VA base Reference-guided generation with manually written prompts Deploy Ref2VA base →

Before you deploy

MiniMax H3 model files are large and depend on ComfyUI offloading. Check the selected GPU, system RAM and persistent volume before clicking Deploy Pod.

GPU profile Tested system RAM Model profile Suitable for
RTX 3090/4090, 24 GB VRAM 50 GB Pruned INT8 ConvRot Lowest-cost compatible option
RTX 5090, 32 GB VRAM 70 GB Pruned INT8 ConvRot Low-VRAM Blackwell option
L40S, 48 GB VRAM 80 GB Full INT8 ConvRot Higher quality and longer video
RTX PRO 6000, 96 GB VRAM 70 GB Full INT8 ConvRot Maximum tested quality and speed

Resolution, duration and concurrent model loading can increase both VRAM and system RAM use. See MiniMax hardware requirements for the tested output limits.

  • Keep the template's persistent volume mounted at /workspace.
  • Do not reduce the template's default storage unless you know the complete model footprint.
  • Set PASSWORD if you want a fixed Code-Server password.
  • Add HF_TOKEN or CIVITAI_TOKEN only when your additional model downloads require them.

See RunPod configuration for all supported environment variables and secret names.

Deploy and connect

  1. Open one of the deployment links above.
  2. Choose an available GPU that matches a tested profile.
  3. Confirm the system RAM and persistent volume settings.
  4. Add any required environment variables or RunPod secrets.
  5. Click Deploy Pod.
  6. Open the pod's logs and wait for the final ready message. Initial startup can take several minutes while the container and models download.
  7. Expand the pod, select Connect, and open ComfyUI under HTTP Services.

For screenshots, startup checkpoints and troubleshooting, follow the full RunPod pod deployment guide.

Container date tags

RunPod templates use dated container tags. Keep the tag supplied by the template unless you are deliberately testing another build. A template can remain on an earlier validated tag when a newer container has not yet been tested with that MiniMax model configuration.

Continue after deployment

Reference

Topic Guide
Hardware and tested output limits MiniMax hardware requirements
Environment variables and network services RunPod configuration
Pod startup and troubleshooting RunPod pod deployment