MiniMax-H3 GGUF (fl2va + ref2va) Q5_1 / Q4_0

Support

Everything here is free and stays free — the format spec, the nodes, the workflows, the cartridges, the LoRAs. If it saved you a night of debugging (it contains several hundred of mine), tips keep the 5090 warm:

• ☕ Buy me a coffee on Ko-fi

• 💛 Sponsor on GitHub

• 🔁 Liberapay (recurring)

• ⚡ Or right here: the Civitai tip button on this page sends Buzz directly.

MiniMax-H3's 33B video+audio DiTs, quantized for 16-32 GB cards. Both flavors — fl2va (text / first-last-frame to video+audio) and ref2va (identity from reference images, videos, and voice clips) — quantized from the Comfy-Org bf16 releases with a first-class minimax_h3 architecture in the metadata.

The files

• fl2va Q5_1 — 25.9 GB · 6.0 bits/weight · the 24-32 GB tier, near-Q6 quality. The demo on the companion workflow page rendered on this file.

• fl2va Q4_0 — 19.9 GB · 4.5 bits/weight · the 16 GB tier (streams the overflow).

• ref2va Q5_1 / Q4_0 — same sizes, for reference-identity workflows.

Coming after validation: Q3mix (17.3 GB) — real 3-bit K-quants on the tensors whose geometry allows it, classic types elsewhere. Named for what it is.

Quick fixes — read this first

• “unknown model architecture: minimax_h3”
  Why: ComfyUI-GGUF doesn't know this architecture out of the box.
  Fix: Install ComfyUI-GGUF, then run the one-line apply_gguf_arch_patch.py from the attached loader pack (idempotent, takes two seconds), restart ComfyUI.

• Where is the text encoder? Where are the VAEs?
  Why: A GGUF here contains the transformer and nothing else.
  Fix: Use the originals from Comfy-Org/MiniMax-H3 (int8 encoder recommended). Load the GGUF with UnetLoaderGGUF — or with the one-dropdown dual-format loader in the companion workflow pack.

• It's 19.9 GB and I have a 16 GB card
  Why: That is the wrong mental model — the file does not need to fit.
  Fix: Nothing. ComfyUI streams what doesn't fit; you pay some speed, not failure. Measured reference: Q5_1 on an RTX 5090 runs ~22 GB resident, about 10 minutes for a 243-frame 544x960 shot at 20 steps.

• Which flavor do I want?
  Why: Two different conditioning systems, two checkpoints.
  Fix: fl2va for text-to-video, image-to-video, and last-frame chaining (the multishot workflow). ref2va when identity comes from reference images or voice clips instead.

Why there is no Q6_K

K-quants (Q6_K, Q5_K, Q4_K, Q3_K…) require weight rows divisible by 256, and H3's hidden width is 2688 — which is not. Requesting one just quantizes something else with the wrong name on it. The ladder for pure quants here is the classic family: Q5_1 and Q4_0 above. One useful exception exists — the FFN down-projections ARE 256-divisible, which is what makes the mixed Q3mix tier possible at 17.3 GB.

Quantization notes

• About 40% of this model is per-block adaln modulation; it is quantized here, exactly as the official int8 release quantizes it.

• Kept at full precision: patch projections, time embedder, condition projection, final layer, token refiner.

• Built with a patched llama.cpp adding minimax_h3 as a first-class architecture — the metadata says what the file actually is.

Everything else I've published

• These weights on Hugging Face

• The multishot workflow on Hugging Face

• Node pack source on GitHub

• All models on Hugging Face