Video models
How to Set Up MiniMax H3 in ComfyUI
A documentation-led setup path for choosing an official template, fetching its files and validating a small first run.
Check versions and choose the input workflow
The official guide reviewed on 26 August 2026 lists ComfyUI 0.30.0+ and native T2V, I2V and R2V templates. Open Template Library → Video, select the matching H3 template, and follow its model-scan prompts. Version requirements may change, so check the current ComfyUI guide before setup.
Download only the files the template requests
The template requests converted files from Comfy-Org/MiniMax-H3. T2V/I2V use FL2VA; R2V requires Ref2VA. Keep the model, text encoder and VAE in the documented locations. For filename roles and variants, see H3 model files. File size alone does not establish that a model will fit in system or video memory.
Validate a small first job
Keep the template defaults and start with a short, low-complexity job. Confirm that the render completes and that the output video and audio behave as expected before increasing resolution or duration. We have not tested this setup on our own hardware, so we do not publish GPU promises or timing estimates.
Common first-run checks
| Symptom | Check |
|---|---|
| Missing model | Confirm the exact filename and folder requested by the selected template. |
| Unsupported operator | Update ComfyUI and the relevant node package, then restart. |
| Memory pressure | Reduce the workflow’s resolution or workload; do not infer capacity from active parameter counts. |
| Unexpected output | Confirm the selected template matches the input type and reference assets. |
For prompt planning see the H3 workflow overview. If local requirements do not fit, compare with the hosted options on the H3 overview.
Sources and version notes
Checked October 23, 2026. We label official claims, third-party measurements, our own testing and community reports separately — see the sources policy for what each label means.