Turn 8-second MiniMax H3 clips into 1-minute+ multi-shot films — same face, same voice, synced dialogue, solo shot retakes, and 2 MP guided upscaling — all from one ComfyUI node.🕹️ <a target="_blank" rel="ugc" href="https://sandpies.github.io/ComfyUI-Hand-Tie-Clips/demo.html">Try the Interactive Node UI in Your Browser</a> (click through Script, Refs, Timeline, Swap &amp; ⚙ RUN before installing) 📦 <a target="_blank" rel="ugc" href="https://github.com/dntpi/ComfyUI-Hand-Tie-Clips">GitHub Repository</a> · <a target="_blank" rel="ugc" href="https://huggingface.co/sandpies/ComfyUI-Hand-Tie-Clips">Hugging Face Mirror</a><hr /><h3 id="what-you-can-do">🎬 What You Can Do</h3><ul><li><code>Simple</code> Mode (Standard MMH3 <code>ref2va</code> / <code>t2va</code> + Realism Boost): Paste standard 6-field MiniMax H3 <code>ref2va</code> / <code>t2va</code> prompts (passed through verbatim!) or plain prose for single-clip generations (<code>chains = 1</code>) with the visual <code>REFS</code> rail, <code>realism_boost: on</code> (<code>steps = 6</code> → 2 base + 10 realism LoRA steps) + <code>audio_refine: on</code> (5 steps), and 2 MP Guided Upsampling!</li><li><code>Shots</code> Mode (Multi-Hop Chains): Write one short beat per card (5 s – 15 s each) with camera, framing, pace, and join controls.</li><li>Visual &amp; Voice Reference Rail (<code>REFS</code>): Drop up to 9 character/location stills and voice timbre clips straight onto the node with <code>@tags</code> and per-hop schedules (no <code>Load Image</code> spaghetti).</li><li>Lossless Retake Board (<code>TIMELINE</code>): Every hop caches losslessly. Lock the shots you like, or click <code>⏵</code> on Shot 4 to re-roll only Shot 4 while Shots 1–3 load in 1 second.</li><li>Standalone Guided <code>SWAP</code>: One-hop character or head swap driven by a reference video's motion &amp; audio (<code>replace_person</code>, <code>head_swap</code>, <code>face_only</code>, <code>keep_person</code>) with a wardrobe override box and auto-loaded Swap LoRAs.</li><li>Built-In 2 MP Guided Upsampler: Enlarges the finished film in 3D latent chunks and runs a 2-step tiled redraw with the <code>h3upscale_v2_x2</code> guide LoRA.</li><li>Local AI Draft Assistant (<code>WRITE</code> — WIP): Connect LM Studio / nInfer / llama.cpp to draft shot cards or bar-synced music video layouts from a 1-sentence brief.</li></ul><hr /><h3 id="two-ways-to-join-shots-(pin_mech)-and-honest-limits">🔗 Two Ways to Join Shots (<code>pin_mech</code>) &amp; Honest Limits</h3><ul><li><code>motion_context</code> (Default — Built-In): Native latent-slice + audio relay (no external custom node needed; credit to <a target="_blank" rel="ugc" href="https://github.com/NikoDemon80/ComfyUI-H3-Motion-Context">NikoDemon80</a>). Honest runway with <code>hop_refine: full</code>: Holds clean for ~5–6 continuous clips on Turbo and ~7–8 clips on non-Turbo before texture changes become visible. Plan a transition to a cut (<code>anchor: "restart"</code>) every 4–5 clips to keep textures fresh across long multi-scene films.</li><li><code>hand_tie</code> (Single-Take Previs — Public Experimental): Renders one low-res single-take bridge previs first and ties each full-res hop's opening and closing 0.9 s to those clean windows so hops never copy texture from each other (10+ hops in one continuous take). Best for continuous walking/talking vlogs in one setting; pair with <code>qwen3vl_4b_fp8_scaled.safetensors</code> in <code>models/text_encoders/</code>.</li></ul><hr /><h3 id="60-second-install">⚡ 60-Second Install</h3>Zero Python/pip dependencies. In <code>ComfyUI/custom_nodes/</code>:<pre><code>git clone https://github.com/dntpi/ComfyUI-Hand-Tie-Clips.git</code></pre><ol><li>Restart ComfyUI and hard-refresh your browser (<code>Ctrl+Shift+R</code>).</li><li>Drag <code>workflows/HandTieClips_Starter.json</code> onto ComfyUI.</li><li>Drop your character photo onto Row 1 in REFS (e.g. <code>@maya</code>), write your beats in SCRIPT, and press Queue.</li></ol>Recommended Models &amp; Companion Nodes:<ul><li>Turbo DiT: <a target="_blank" rel="ugc" href="https://huggingface.co/sandpies/Wuli-MiniMax-H3-Turbo-ComfyUI">Wuli MiniMax H3 8-Step Turbo</a> + plain <code>minimax_h3_ref2va</code> on <code>refine_model</code> / <code>upsample_dit</code></li><li>Guided Upscaler: <code>minimax_h3_latent_upscaler_3d_bf16.safetensors</code> (<code>models/latent_upscale_models/</code>) + <code>h3upscale_v2_x2.safetensors</code> (<code>models/loras/</code>)</li><li>Speed / Attention Nodes: <a target="_blank" rel="ugc" href="https://github.com/PlagueKind/ComfyUI-PlagueKind-Nodes">ComfyUI-PlagueKind-Nodes</a> &amp; <a target="_blank" rel="ugc" href="https://github.com/kijai/ComfyUI-KJNodes">ComfyUI-KJNodes</a></li></ul><hr /><h3 id="the-5-golden-prompting-rules-(h3-runs-at-cfg-1.0-every-word-is-additive)">💡 The 5 Golden Prompting Rules (H3 runs at CFG 1.0 — every word is additive)</h3><ol><li>Never name what you want to stop: Don't write "She stops talking" — write "She smiles with her lips closed. Gentle pool water lapping."</li><li>Silence must be written as a room tone: An empty soundtrack invents speech. Always name a narrowband sound ("low hum of the refrigerator", "soft breeze").</li><li>Don't re-describe the face on Shots 2+: Describe only the new action; the reference rail and join pin carry her look.</li><li>No ellipses (<code>...</code>) inside dialogue: Connect clauses inside speech with a simple comma (<code>,</code>) so the audio branch never fills pause gaps with gibberish.</li><li>Set the final shot's <code>tail</code> to <code>settle</code> or <code>hold</code>: So the film lands cleanly instead of starting a new gesture on the last frame.</li></ol>

Hand Tie Clips — multi-hop MiniMax H3 video + audio, one queue

What's new in 2.2.0 — 2026-09-21Three corrections to the refine pass shipped in 2.1.0, plus the MEDIA audio window.<code>refine_align</code> picks the sigmas the refine pass re-runs. The default stays <code>denoise</code> — a 2.1 graph reloads on the schedule it rendered under — and <code>speed_mode=turbo</code> now sets <code>hop_tail</code>, which re-runs the hop's own last steps instead of inventing a schedule tuned for a 4-step base.<code>refine_blend=auto</code> derives the blend ramp from this run's <code>overlap</code>. The old literal <code>0:0, 22:0, 44:1</code> had the overlap width baked into it, so at a 1.6 s overlap it refined frames inside the pin — exactly the seam cost the ramp exists to remove — and at 0.2 s it threw away refined frames you had paid for. <code>auto</code> at the shipped 0.9 s overlap resolves to the old string byte-for-byte, so cached hops still hit.SLA moved to the front of the MODEL wire in both shipped workflows, on PlagueKind's recommendation. That reorder changes <code>_model_fingerprint</code>, so every hop re-renders once on first run even with <code>cache_hops=on</code>.Also: <code>master_audio_start_s</code>, so the chain's audio window can start somewhere other than zero.Full notes stay at the bottom under <a target="_blank" rel="ugc" href="https://github.com/dntpi/ComfyUI-Hand-Tie-Clips#changelog">Changelog</a>. Detail lives in <code>CHANGELOG.md</code>.

Files with SHA256: 7e1743c79ca3fde8c10b4f1d597d4858a5dda2868747411e62eb2c839b9cb402