Skip to content

🚀 Template Deployment RunPod

  • The MiniMax image is intended for NVIDIA CUDA GPUs.

🚀 Template

Template Tasks Inputs Output
MiniMax H3 FL2VA Text-to-video, image-to-video and first/last-frame-to-video Text with optional first and/or last frame Video with audio
MiniMax H3 Ref2VA Reference-to-video Text with reference images, video and/or audio Video with audio

Two prompt-enhancement options are available

  • The tail uses the MiniMax H3 text encoder and is slower.
  • The QWEN prompt enhancer uses a second, separate Qwen model together with llama.cpp for faster prompt generation.
  • Both are only needed for prompt enhancement, not for inference with the MiniMax H3 model itself.

🏷️ Container date tags

The dated image tag used by a RunPod template can remain on an earlier validated build when a newer image has not yet been tested for this model.

📘 Tutorial

💻 Hardware requirements

⚙️ Configuration