<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><h2 id="identity-from-reference-images-video-and-audio-instead-of-a-start-frame.">Identity from reference images, video and audio instead of a start frame. </h2>Point it at photographs of a person, a place, a voice — up to 9 images, 3 videos, 3 soundtracks and 3 standalone audio clips — and the render carries them.Three things you need besides this download. Both of the support questions this pack generated on day one were setup, not bugs, so they are first:<ul><li>The node pack — <a target="_blank" rel="ugc" href="https://github.com/jlucasmcrell/ComfyUI-H3-Multishot">ComfyUI-H3-Multishot</a>. Install via Git URL, restart, then hard-refresh the browser tab; the frontend caches node definitions and you will otherwise see red nodes. Then run <code>python apply_gguf_arch_patch.py</code> from its folder and restart again. Without that patch, loading any H3 GGUF fails with <code>ValueError: Unexpected architecture type in GGUF file: 'minimax_h3'</code> — ComfyUI-GGUF keeps a hardcoded list of architectures and this one is not on it.</li><li>A ref2va DiT — <a target="_blank" rel="ugc" href="https://civarchive.com/models/2835678">curve form</a> (smaller, needs ComfyUI 0.30.0+) or <a target="_blank" rel="ugc" href="https://civarchive.com/models/2833352">original form</a>. <code>fl2va</code> files will not work in this mode.</li><li>The text encoder and VAEs — <a target="_blank" rel="ugc" href="https://civarchive.com/models/2834385">Qwen3-VL encoder</a>, and take the <code>mmproj</code> sidecar with it. Load it with the pack's <code>H3ClipLoaderAny</code> at type = minimax, not with "CLIPLoader (GGUF)" — that node has no minimax type, falls back silently, and dies deep inside a tokenizer with an unrelated-looking JSON error.</li></ul><h2 id="naming-your-references-in-the-prompt">Naming your references in the prompt</h2>This is the part that is not guessable. Reference blocks are labelled in the prompt text, and the labels are 1-based while the input slots are 0-based — so <code>ref_image_0</code> is <code>&lt;Picture 1&gt;</code>.<pre><code>&lt;Picture 1&gt; is the woman. &lt;Picture 2&gt; is the room she is standing in.
She walks to the window in the style of &lt;Video 1&gt;.</code></pre>Ordering is fixed by the node regardless of how you wire it: all images, then videos, then standalone audios. Empty slots are skipped and the numbering compacts.The ordinal trap. A reference video with a soundtrack consumes an <code>&lt;Audio&gt;</code> ordinal before your standalone clips. One ref video with sound plus one <code>ref_audio_0</code> makes your voice clip <code>&lt;Audio 2&gt;</code>, not <code>&lt;Audio 1&gt;</code>. Get it wrong and the wrong voice binds, with no error.<h2 id="references-and-keyframes-cannot-be-combined">References and keyframes cannot be combined</h2>Not a design choice here — a hard limit in ComfyUI's core. <code>model_base.py</code> writes <code>cond_video_latents</code> for keyframes, then assigns it again for references, so the keyframe latents are discarded while the keyframe layout rows survive. The packed sequence desyncs and the sampler dies on a shape mismatch. There is no "reference images plus start frame" mode, and asking for both crashes rather than degrading. Use the keyframes graph when you need a specific opening or closing frame.<h2 id="settings-measured-rather-than-guessed">Settings, measured rather than guessed</h2><ul><li>Leave the eviction node ON. It frees the ~16.5 GB text encoder before the DiT loads. Images only: 3.4 min with it against 4.5 without, and the output is bit-identical — every pairing scored 0.00 mean pixel distance. With a reference video it is the difference between 6.5 min and a 46-minute run that was killed.</li><li>Use <code>ref_image_size = match</code>, not <code>max</code>. Same subject, same seed: <code>max</code> cost 24% more time and rendered a facial scar as a vague smudge on the wrong eyebrow, with less fine detail. It does not buy fidelity.</li><li>Prefer more references over bigger ones. Adding a third reference image cost nothing measurable — 2.82 min against 2.87.</li><li>Reference audio must be STEREO. The audio VAE encodes <code>[B, 2, L]</code> and a mono clip produces half the rows it reserved, then dies deep in the model with no useful error. The stereo-guard node handles it; keep it wired. Audio costs about 14% wall clock and trims motion ~14–27%, which is the conditioning constraining the performance rather than stalling it.</li><li>Reference video rules: minimum 5 frames, silently truncated to the generation length then trimmed to <code>n % 17 == 5</code>, and shown to the encoder at 2 fps.</li></ul><h2 id="if-a-render-crawls">If a render crawls</h2>Check power draw, not utilisation. A card thrashing weights between system RAM and VRAM still reports ~98% utilisation while pulling a fraction of its rated watts. The tell in your log is:<pre><code>loaded partially; 6423 MB usable, 5847 MB loaded, 19363 MB offloaded</code></pre>Sample the wattage for 20–30 seconds before concluding anything. The encoder/conditioning phase at the start of every render is legitimately quiet — around 100 W at ~35% utilisation — and one unlucky reading there looks exactly like thrashing. If it climbs past 400 W once sampling starts, nothing is wrong.<h2 id="what-is-in-the-download">What is in the download</h2><ul><li><code>H3_HardMode_R2V.json</code> — the reference-to-video graph, every node labelled</li><li><code>workflows/</code> — the multishot AIO, long-form memory and keyframes graphs. These are also in the MiniMax-H3 Multishot pack you may already have; the copies here are relabelled so nothing shows a raw class name.</li><li><code>reference_images/</code> — three views of one synthetic subject, so the graph works the moment you load it rather than pointing at ComfyUI's <code>example.png</code></li><li><code>HARD_MODE_GUIDE.html</code> — the full guide, including the parts above</li></ul>The reference images are synthetic, generated for this release, and free to use. Weights are under MiniMax's own licence; the workflows and node pack are MIT.

Files with SHA256: 6f989c084689e0461f3f3d56f593af4668397f908ce1d3d67668454586d72429