The useful answer

  • Begin with a ComfyUI build that includes native H3 support.
  • H3 uses shared components plus task-specific checkpoints.
  • Validate one simple 768p workflow before adding optimizations.
  • Quantization can reduce memory pressure but changes the setup path.

Start with the supported runtime

MiniMax H3 support depends on H3-aware nodes and model loaders. Update ComfyUI through its supported update path, restart it, and confirm that H3 model types appear before downloading a large checkpoint collection.

Avoid mixing an old custom node, a newer native node and an unrelated workflow JSON. Each may expect a different folder layout or tensor format. Pick one documented path and follow it end to end.

Understand the model pieces

The official release provides two task-oriented checkpoints: FL2VA for text or first/last-frame generation, and Ref2VA for ordered reference inputs. Shared tokenizer, processor, text encoder, visual VAE and audio VAE components are also required.

This is why an H3 install is not equivalent to dropping one checkpoint file into a generic models folder. Use the directory structure described by the workflow author and keep the official model index files with the download.

  • FL2VA: text-to-audio-video and first/last-frame conditioning.
  • Ref2VA: reference image, video and audio conditioning.
  • Shared assets: encoders, tokenizers, processors and VAEs.
  • Optional adapters: acceleration or style layers with separate compatibility requirements.

Run a minimal validation

Use a small, known workflow and a short prompt. Keep references out of the first test. Confirm that the model loads, video frames decode, stereo audio is present and the final MP4 is playable.

If loading fails, inspect the first error rather than changing several settings at once. Folder mismatch, incomplete download, incompatible node version and insufficient memory are separate problems and need different fixes.

Hosted versus local cost

Local generation replaces a per-second API invoice with hardware time, electricity, setup and maintenance. If H3 is an occasional tool, a hosted endpoint can be cheaper. If the machine is already available and utilization is high, local control may justify the setup cost.

Questions answered

What people ask about minimax h3 comfyui

Does ComfyUI support MiniMax H3?

Yes. H3 can run through supported ComfyUI workflows, including native model support and community templates.

Which H3 checkpoint should I use in ComfyUI?

Use FL2VA for text and first/last-frame workflows, and Ref2VA for ordered reference-media workflows.

Should I begin with GGUF or Turbo LoRA?

Begin with the simplest compatible baseline your hardware can run, then add one optimization at a time so failures remain diagnosable.

Primary reading

Sources behind this guide

MiniMax H3 official repositoryReference implementation, checkpoints and prompt guidance.MiniMax H3 official model cardWeights, architecture, capabilities and license links.MiniMax H3 integration directoryOfficially maintained map of runtimes, training tools and community projects.