<h2 id="support">Support</h2>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:<ul><li>☕ <a target="_blank" rel="ugc" href="https://ko-fi.com/joeygambino">Buy me a coffee on Ko-fi</a></li><li>💛 <a target="_blank" rel="ugc" href="https://github.com/sponsors/jlucasmcrell">Sponsor on GitHub</a></li><li>🔁 <a target="_blank" rel="ugc" href="https://liberapay.com/joeygambino">Liberapay</a> (recurring)</li><li>⚡ Or right here: the Civitai tip button on this page sends Buzz directly.</li></ul>MiniMax-H3's 33B video+audio DiTs, quantized for 16-32 GB cards. Both flavors — <code>fl2va</code> (text / first-last-frame to video+audio) and <code>ref2va</code> (identity from reference images, videos, and voice clips) — quantized from the Comfy-Org bf16 releases with a first-class <code>minimax_h3</code> architecture in the metadata.<h2 id="the-files">The files</h2><ul><li>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.</li><li>fl2va Q4_0 — 19.9 GB · 4.5 bits/weight · the 16 GB tier (streams the overflow).</li><li>ref2va Q5_1 / Q4_0 — same sizes, for reference-identity workflows.</li></ul>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.<h2 id="quick-fixes-read-this-first">Quick fixes — read this first</h2><ul><li>“unknown model architecture: minimax_h3” Why: ComfyUI-GGUF doesn't know this architecture out of the box. Fix: Install <a target="_blank" rel="ugc" href="https://github.com/city96/ComfyUI-GGUF">ComfyUI-GGUF</a>, then run the one-line <code>apply_gguf_arch_patch.py</code> from the attached loader pack (idempotent, takes two seconds), restart ComfyUI.</li><li>Where is the text encoder? Where are the VAEs? Why: A GGUF here contains the transformer and nothing else. Fix: Use the originals from <a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/MiniMax-H3">Comfy-Org/MiniMax-H3</a> (int8 encoder recommended). Load the GGUF with <code>UnetLoaderGGUF</code> — or with the one-dropdown dual-format loader in the companion workflow pack.</li><li>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.</li><li>Which flavor do I want? Why: Two different conditioning systems, two checkpoints. Fix: <code>fl2va</code> for text-to-video, image-to-video, and last-frame chaining (the multishot workflow). <code>ref2va</code> when identity comes from reference images or voice clips instead.</li></ul><h2 id="why-there-is-no-q6_k">Why there is no Q6_K</h2>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.<h2 id="quantization-notes">Quantization notes</h2><ul><li>About 40% of this model is per-block adaln modulation; it is quantized here, exactly as the official int8 release quantizes it.</li><li>Kept at full precision: patch projections, time embedder, condition projection, final layer, token refiner.</li><li>Built with a patched llama.cpp adding <code>minimax_h3</code> as a first-class architecture — the metadata says what the file actually is.</li></ul><h2 id="everything-else-i've-published">Everything else I've published</h2><ul><li><a target="_blank" rel="ugc" href="https://huggingface.co/joeygambino/MiniMax-H3-GGUF">These weights on Hugging Face</a></li><li><a target="_blank" rel="ugc" href="https://huggingface.co/joeygambino/MiniMax-H3-Multishot-Workflow">The multishot workflow on Hugging Face</a></li><li><a target="_blank" rel="ugc" href="https://github.com/jlucasmcrell/ComfyUI-H3-Multishot">Node pack source on GitHub</a></li><li><a target="_blank" rel="ugc" href="https://huggingface.co/joeygambino">All models on Hugging Face</a></li></ul>

Files with SHA256: 9d93f666cf24fbfc34305bd9fb4ed6e14f408d06b22544d8a075f2c48790cea8