Turn 8-second MiniMax H3 clips into 1-minute+ films — same face, same voice, synced dialogue, one-shot retakes, one continuous take with no joins, and 2 MP 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/sandpies/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><h2 id="new-in-3.4.0:-one-continuous-take-(sand_tie)">🆕 New in 3.4.0: One Continuous Take (<code>sand_tie</code>)</h2><ul><li>A whole run of hops renders as ONE video. No joins to see; a take only cuts where you write <code>anchor: restart</code>.</li><li>🎬 <a target="_blank" rel="ugc" href="https://media.githubusercontent.com/media/sandpies/ComfyUI-Hand-Tie-Clips/main/docs/img/sand-tie-mv-walk-me-through-the-night.mp4">2:03 one-take music video</a> · <a target="_blank" rel="ugc" href="https://media.githubusercontent.com/media/sandpies/ComfyUI-Hand-Tie-Clips/main/docs/img/sand-tie-two-person-place-change-52s.mp4">52 s two-person take, street → boardwalk → bench</a> · <a target="_blank" rel="ugc" href="https://media.githubusercontent.com/media/sandpies/ComfyUI-Hand-Tie-Clips/main/docs/img/sand-tie-walking-selfie-65s.mp4">65 s walking selfie (upsampled)</a></li><li>Retake one hop and only that hop is redrawn; the rest of the take is kept.</li><li><code>tone_lock</code> keeps a long take from drifting brighter (0.5 / 0.75 / full).</li><li>Song-locked takes (<code>master_audio_file</code>) keep lip sync for the whole song.</li></ul><h2 id="what-you-can-do">🎬 What You Can Do</h2><ul><li>Simple Mode (Standard MMH3 ref2va / t2va + Realism Boost): Paste standard 6-field MiniMax H3 prompts (passed through verbatim!) or plain prose for single clips (<code>chains = 1</code>) with the visual REFS 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 upsampling.</li><li>Shots Mode (Multi-Hop Chains): One short beat per card (5 s – 15 s each) with camera, framing, pace and join controls.</li><li>Visual &amp; Voice Reference Rail (REFS): Up to 9 character/location stills and voice clips right on the node, with @tags and per-hop schedules (no Load Image spaghetti).</li><li>Lossless Retake Board (TIMELINE): Every hop is cached. Lock the shots you like, or click ⏵ on Shot 4 to re-roll only Shot 4 while Shots 1–3 load in a second.</li><li>Guided SWAP: 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 Upsampler: Enlarges the finished film with the H3 latent upscaler and a light 3-step Open-H3-Regenerate-2K redraw. Small, distant faces are redrawn as close-ups so they keep their person.</li><li>WRITE (Local AI Writer): Connect LM Studio / llama.cpp to draft shot cards, one-take timelines or music video layouts from a one-sentence brief. STANDARD writes a single checked H3 prompt.</li></ul><h2 id="three-ways-to-join-shots-(pin_mech)">🔗 Three Ways to Join Shots (<code>pin_mech</code>)</h2><ul><li><code>motion_context</code> (Default — Built-In): Native latent-slice + audio relay (credit to NikoDemon80). Honest runway with <code>hop_refine: full</code>: ~5–6 clips on Turbo, ~7–8 on non-Turbo before texture changes show. Plan a cut (<code>anchor: "restart"</code>) every 4–5 clips.</li><li><code>sand_tie</code> (New — One Continuous Take): Each take renders as one video, its hops overlapping windows. Best for long walks, songs and scenes that move through places. Needs a recent ComfyUI (MiniMax H3 context windows) and <code>overlap</code> 0.9 s.</li><li><code>hand_tie</code> (Experimental): Renders a low-res preview of the take first and ties every hop to it. Pair with <code>qwen3vl_4b_fp8_scaled.safetensors</code> in <code>models/text_encoders/</code>.</li></ul><h2 id="60-second-install">⚡ 60-Second Install</h2>Zero pip dependencies. Needs ComfyUI v0.38.0 or newer. In <code>ComfyUI/custom_nodes/</code>:<code>git clone https://github.com/sandpies/ComfyUI-Hand-Tie-Clips.git</code><ol><li>Restart ComfyUI and hard-refresh your browser (Ctrl+Shift+R).</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><h3 id="recommended-models-and-companion-nodes">🧩 Recommended Models &amp; Companion Nodes</h3><ul><li>Turbo DiT: Wuli MiniMax H3 8-Step Turbo + plain <code>minimax_h3_ref2va</code> on <code>refine_model</code> / <code>upsample_dit</code></li><li>Upsampler: <code>minimax_h3_latent_upscaler_3d_bf16.safetensors</code> (<code>models/latent_upscale_models/</code>) + <code>open_h3_regenerate_2k_r256pp_bf16.safetensors</code> (<code>models/loras/</code>)</li><li>Speed / Attention Nodes: ComfyUI-PlagueKind-Nodes &amp; ComfyUI-KJNodes</li></ul><h2 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)</h2><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 carry her look.</li><li>Only words inside the quotes: A laugh or sigh written inside a line is said out loud as a word. Write it outside: she laughs, then says, '…'</li><li>Set the final shot's tail 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>🎞️ One-take tip: in a <code>sand_tie</code> take, every hop opens with the same camera sentence and the action goes on an <code>At mm:ss-mm:ss,</code> timeline. PROMPTING.md has the full rules; WRITE does it for you.

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

3.4.0: one continuous take- New pin_mech: sand_tie: a whole run of hops renders as one video, so there are no joins to see; a take only cuts at anchor: restart. Tested on a 2-minute song-locked MV, a two-person scene that changes place, and walking selfies. It needs a recent ComfyUI.- Retake one hop: only that hop is redrawn, and the rest of the take is kept.- tone_lock: stops long takes drifting brighter (0.5 / 0.75 / full).- Upsampler: small, distant faces are redrawn as close-ups so they keep their person, and long clips no longer run out of RAM.- Writer: new rules and notes for lines said together, sounds written inside a line, talking while eating, and camera arcs.- Tooltips: every input now has a short, plain one.Saved graphs still load. The full list is in <a target="_blank" rel="ugc" href="https://github.com/sandpies/ComfyUI-Hand-Tie-Clips/blob/main/CHANGELOG.md">CHANGELOG.md</a><a target="_blank" rel="ugc" href="https://github.com/sandpies/ComfyUI-Hand-Tie-Clips/blob/main/CHANGELOG.md">.</a>

Files with SHA256: 00f37cc4f040a724f501d7af7e48aea9a774fd5c5132c61b305e139ff2fb6c38