Hand Tie Clips — Multi-Shot Studio for MiniMax H3

Multi-hop MiniMax H3 video from one ComfyUI node. Author shot cards, manage reference stills and voices on a visual rail, lock or solo-retake individual hops from a lossless cache, run guided character/head swaps, and upscale to 2 MP.

✓ Zero Pip Dependencies ✓ Built-In Motion-Context & Hand-Tie ✓ Classic LiteGraph & Nodes 2.0 ✓ Lossless Hop Cache & Solo Retake

Six consecutive frames crossing the join between Hop 1 and Hop 2 — no cut, flash, or identity jump.

Hand-Tie 10x8s example

Interactive Node Preview — Click the Tabs & ⚙ RUN

H3 Ref2VA Chain (HandTieClips)

✨ Interactive Demo — Click any tab, lock button, or ⚙ RUN

model clip vae audio_vae refine_model upsample_dit audio_model

images audio info contact_sheet

768×1344 (9:16) · 8s ×4 (29.2s) · 8 steps euler/simple · pin: motion_context · cache on ● Ready

SCRIPT 10 shots · 71.8s master

NEXT QUEUE re-renders none · 10 cached

JSON

✨ Simple Mode (Standard MMH3 ref2va / t2va Studio + Realism Boost): Paste any standard 6-field MiniMax H3 ref2va / t2va prompt (subject_definitions:, summary:, detailed_description:, etc.) or plain prose directly here with chains = 1. HTC passes structured H3 prompts straight through untouched while giving you the visual REFS rail, realism_boost: on (steps = 6 → 2 base + 10 realism LoRA steps) + audio_refine: on (5 steps), and 2 MP Guided Upsampling!

STANDARD MMH3 6-FIELD PROMPT (ref2va / t2va) OR PLAIN PROSE chains = 1 · 15.0s (362f) · steps = 6 (2 + 10 realism_boost) · audio_refine = on · upsample = on

subject_definitions:
<Subject 1> is the young woman in <Picture 1>. The collage grid is identity only. She has dark hair in an updo with loose face-framing strands and small flower clips, a white triangle bikini with blue flowers, blue string ties at the neck and hips, and bare feet.
<Subject 2> is the nugget mascot in <Picture 2>, an orange breaded body, blue headband, yellow beak, round black eyes, small wings, and two feet.
<Subject 3> is the beach in <Picture 3>: raked sand, blue court lines, a blue-padded volleyball net, palm trunks, and the ocean horizon.

summary:
[reference generation] <Subject 1> dances a fast body roll and hip circle on the kick at the center of <Subject 3> after dark, with a lighting truss, speaker stacks, a hexagonal neon panel, a full moon, string lights, and a crowd ring around the court. <Subject 2> bounces beside her on that same kick. Four cuts only change the angle. The camera arcs and pushes in, rolls low along the hip circle, pedestals into a hero view while her hands travel up the torso, then tracks her turn into a close call-me hand.

retention_analysis:
<Subject 1> (appears in [Shot 1], [Shot 2], [Shot 3], [Shot 4]): fully_preserved - her face, updo, flower clips, white floral bikini, blue string ties, and bare feet are retained.
<Subject 2> (appears in [Shot 1], [Shot 2], [Shot 3], [Shot 4]): fully_preserved - the breaded orange body, blue headband, yellow beak, black eyes, small wings, and two feet are retained.
<Subject 3> (appears in [Shot 1], [Shot 2], [Shot 3], [Shot 4]): partially_preserved - raked sand, blue court lines, the blue-padded net, palm trunks, and the ocean horizon stay; daylight becomes night and a stage, moon, string lights, and crowd occupy the court.

detailed_description:
Photoreal live-action night beach. <Subject 1> and <Subject 2> keep the rendered look of their reference images under practical neon and moonlight.
[Shot 1] A high wide shot looks over a glow-stick crowd toward the sand. <Subject 3> shows raked sand, blue court lines, and the blue-padded net at frame-left, with palm trunks, the ocean, a full moon, a lighting truss, speaker stacks, and a hexagonal neon panel behind the pair. <Subject 1> is already mid-roll at center, weight in the hips, bare feet planted. <Subject 2> is at her camera-right side, wings already lifting on the kick.
At 00:00.000-00:01.400, on each kick her hips circle frame-right and the ribcage follows a half-beat later. The updo sways with the circle and the blue hip ties swing. The camera arcs right with large amplitude at fast speed and pushes in. <Subject 2> bounces once per kick, wings flicking at the top of each bounce.
At 00:01.400-00:02.900, the circle reverses frame-left on the next kicks. Her chest rolls up through the circle, shoulders loose, and both hands rest on the hip ties, fingers following the sway. Her mouth is open in a laugh toward the lens. The arc and push keep moving.
At 00:02.900-00:04.400, the hips return frame-right and the chest finishes the roll high. Loose strands cross her cheek on the sway. The camera arrives at an eye-level full-body view, still pushing in with medium amplitude.
[Shot 2] At 00:04.400, the camera cuts low. The hip circle is already traveling frame-right, and both bare feet stay planted while the knees soften on the kick.
At 00:04.400-00:06.100, the camera arcs left with large amplitude at fast speed along her hips and thighs. Each kick drops one hip and lifts the other, the ties flicking at the top of the circle. A little sand shifts under the balls of her feet. <Subject 2> bounces at frame-right, wings marking the kick.
At 00:06.100-00:08.200, the camera rolls counterclockwise and keeps the arc. Her left hip leads a deeper circle toward <Subject 2>, the torso folding with it and the chest already rolling back up. <Subject 2> leans in on the bounce, beak near her shoulder, wings spread.
[Shot 3] At 00:08.200, the camera cuts to a view just behind <Subject 2>. Her chest is already rising out of that fold, and her right hand is already leaving the hip tie.
At 00:08.200-00:10.000, the camera pedestals up with large amplitude at fast speed. On the kicks her right hand travels from the hip tie up the side of the ribcage to the collarbone, wrist loose, while the hips keep the circle. The updo lags a beat behind the chest. <Subject 2> turns toward the lens on the bounce, wings wide.
At 00:10.000-00:11.800, the camera arcs right with large amplitude at fast speed into a hero two-shot and pushes in. Her left hand joins the path and both arms open from the collarbone toward the lens, still riding the hip circle. Her eyes meet the lens, mouth open in the laugh. <Subject 2> faces the lens at frame-left, bouncing on the kick.
[Shot 4] At 00:11.800, the camera cuts high behind the pair. The open arms and the hip circle are already carrying her into a left turn. The crowd is a semicircle. The net, palms, and ocean of <Subject 3> read past the stage.
At 00:11.800-00:13.400, the camera tracks that turn with large amplitude at fast speed. Her hips lead on the kick, shoulders and updo follow, and the ties whip after the turn. <Subject 2> bounces through the same turn and lands facing the lens. Her face comes around to the lens before the camera finishes the arc.
At 00:13.400-00:15.080, the camera pushes in with medium amplitude at fast speed to a close view, <Subject 2>'s headband and one wing remaining at frame-left. The hip circle continues small under the frame. Her right hand, still traveling from the collarbone path, arrives beside her cheek, thumb near the ear and little finger toward the chin. The updo keeps the sway through the last kick.

overall_soundscape:
Crowd roar and a soft shift of sand under bare feet continue under the stage.

non_diegetic_music:
Fast club techno. A distorted four-on-the-floor kick, a rolling low bass, closed metallic hi-hats, and a short minor synth stab. The tempo stays fast and the level stays loud from 00:00.000 through 00:15.080. Her hip circle lands on the kick.

Detects standard 6-field H3 headers automatically and passes them through verbatim while binding REFS rail photos & voices.

REFERENCE REGISTER · STABLE @TAGS & PER-REF PIXEL BUDGETS

<Pic 1>

<Pic 2>

<Pic 3>

✓ NEXT QUEUE: 0 of 4 hops will re-render — all 4 hops served from lossless FFV1 + safetensors cache.

START IMAGE & MASTER AUDIO LOCK

Start Image (Optional Frame 0)

Master Audio File (Continuous Lip-Sync Take)

REFERENCE CLIPS (1–3) & VOICES (1–3)

Reference Clip 1 (<Video 1>)

Voice 1 (<Audio 1>)

WRITE — LOCAL AI PLAN DRAFTING ASSISTANT (WIP) ● nInfer · Qwen3.8-27B — Ready

Drafts cards + reference descriptions. Always review generated beats before rendering.

GUIDED VIDEO SWAP — STANDALONE ONE-HOP IDENTITY / HEAD SWAP Auto-loads Character Swap + Head Swap LoRAs

Identity Photo (@tag)

Wardrobe Override (Optional Free Text)

Compiled Swap Prompt Preview (Written to swap_script, leaving your Shot Plan untouched)

RUN 48 dials · 9 groups (obsolete v1 band-aids removed)

output & sampling

resolution

duration

steps

sampler / scheduler

join & pin (cleaned — 5 essential controls)

pin mech

ref image size

audio pin frames

voice every hop

last frame guide

refine · realism boost · upsample · soundtrack (audio refine)

hop refine

realism boost

audio refine (pair w/ boost)

upsample (h3upscale x2)

Simple Mode (15s · Realism Boost + 2 MP Upsample) example

Two Ways to Join Shots (`pin_mech`)

Mode	Status & Best Use Case	How It Works & Honest Tradeoffs
`motion_context` (Default — Built-In)	Stable Continuous relays (~5–6 hops on Turbo, ~7–8 hops on non-Turbo with Refine) & multi-scene `restart` cuts	Built directly into Hand Tie Clips (no external custom node needed; credit to NikoDemon80's ComfyUI-H3-Motion-Context for the latent-slice + end-aligned audio timeline concept). Relays the previous hop's sampler AV latent into the next hop with in-latent head inpainting and distribution locking — and resets cleanly on `restart` cuts.
`hand_tie` (Single-Take Previs)	Public Experimental Continuous one-location walking vlogs & tracking shots	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. Previs Compiler Model (Recommended in `ComfyUI/models/text_encoders/`): Drop any `qwen3vl_4b.safetensors` or `qwen3vl_8b.safetensors` into `models/text_encoders/` — e.g. `qwen3vl_4b_fp8_scaled.safetensors` (Comfy-Org), `qwen3vl_4b` (silveroxides/K2Q), or `qwen3vl_8b_int8_convrot.safetensors` (Comfy-Org). It loads at the start of the run to compile your beats into continuous bridge moments and unloads from VRAM before H3 starts. Caveat: Because every hop must meet its bridge window, beats that ask for a sudden location change, hard camera angle jump, or fast prop action across a join can cause the previs to cut or morph. It works very well when the camera and subject flow continuously in one setting, and may occasionally need a seed re-roll if a beat moves too far.

Built-In Finishing (`⚙ RUN` Page)

1. Realism Boost + Audio Refine (Built for `Simple` Mode / 1–2 Hop Hero Clips)

Using HTC for standard single generations (ref2va / t2va): You don't need a multi-hop chain to use Hand Tie Clips — switch SCRIPT to Simple mode (chains = 1 in ⚙ RUN) and paste your standard 6-field MiniMax H3 ref2va / t2va prompt (subject_definitions:, summary:, detailed_description:, etc.) or plain prose. HTC passes structured H3 prompts straight through untouched while giving you the visual REFS rail, realism_boost, and upsample on a single clip!
realism_boost: on runs a 2-stage split schedule inside the sampler (at steps = 6: 2 base steps set composition, then 10 realism LoRA steps paint skin and lighting with no fresh noise).
Important Audio Note: Because stage 2 adds no noise, the audio stream only gets those first 2 noisy steps and will sound thin or under-converged on its own. When using realism_boost, turn on audio_refine: on (under the Soundtrack group, 4–5 steps) — it re-samples the soundtrack cleanly while keeping your finished video 100% frozen.

2. Refine Pass (`hop_refine: full`) — Honest Limitations

Runs a gentle low-sigma detail pass per hop (supports wiring a stock 30-step H3 DiT to refine_model while your main wire runs a Turbo model).
The honest truth on chain length: Refine holds back relay degradation and keeps joins clean, pushing continuous motion_context chains to 5–6 clips on Turbo before very visible texture changes appear, and 7–8 clips on non-Turbo. Under motion_context, plan to transition to a cut (anchor: "restart") every 4–5 clips or so to maintain the integrity of your textures (or use pin_mech: hand_tie for continuous single-take sequences).

3. Guided Tiled Upsampler (`upsample: on`) & Standalone Node

Enlarges the finished film to 1.5–2.0 MP in latent chunks and runs a 2-step tiled redraw. With the h3upscale x2 LoRA, the redraw is guided by your actual canvas frames at half the output size so it restores real detail instead of inventing texture. Toggling upsample on a cached chain skips hop generation and runs only the upscaler.

Required Files for Upsampling:

Latent Upscaler (`ComfyUI/models/latent_upscale_models/`):
Download minimax_h3_latent_upscaler_3d_bf16.safetensors from LBH-123-AI/Minimax_h3_latent_Upscaler (or h3_upscaler_lms_v0.1.safetensors).
Guided Upscale LoRA (`ComfyUI/models/loras/`):
Download h3upscale_v2_x2.safetensors from sandpies/ComfyUI-H3-Video-Upsampler (Hugging Face).
upsample_dit wire: When using h3upscale_v2_x2, wire a plain, unpatched H3 UNETLoader into upsample_dit (the LoRA was trained on the bare DiT without turbo/realism LoRAs underneath).

Want to upscale any existing video outside a chain? Use the included H3 Video Upsampler (finished clip) node (workflows/HandTieClips_Upsample_Video.json) or the standalone repo: dntpi/ComfyUI-H3-Video-Upsampler.

Quick Start (60 Seconds)

cd ComfyUI/custom_nodes
git clone https://github.com/dntpi/ComfyUI-Hand-Tie-Clips.git

Restart ComfyUI and hard-refresh your browser (Ctrl+Shift+R). Zero pip dependencies.
Drag workflows/HandTieClips_Starter.json onto ComfyUI and point the loaders at your MiniMax H3 ref2va checkpoint, text encoder, and VAEs.
In REFS, drop your character photo onto Row 1 and give it a tag like @maya.
In SCRIPT, write one beat per shot card and press Queue.

Prompting Cheat-Sheet (The 5 Golden Rules)

MiniMax H3 samples at CFG 1.0 with no negative prompt — every word you write is added to the scene.

❌ Don't Write	✅ Write Instead
"She stops talking." (Names talking → keeps her talking)	"She smiles with her lips closed. The low hum of the refrigerator."
"No background music, silent room." (Silence = model invents speech)	Always name a specific room tone: "Soft rain on the windowpane, distant traffic hum."
Re-describing hair, face, and clothes on Shots 2, 3, 4	Describe only the new action on Shots 2+ — the reference rail and join pin carry her look.
Ending Shot 2 mid-sentence when Shot 3 should be quiet	Finish the action/sentence at the end of Shot 2 so Shot 3 inherits a settled state.
Leaving the final shot's `tail` on `ongoing`	Set the last shot's `tail` to `settle` or `hold` so the film lands cleanly.

📖 Full authoring guide: PROMPTING.md · Release history: CHANGELOG.md · Credits & Notices: THIRD_PARTY_NOTICES.md

Hand Tie Clips — Multi-Shot Studio for MiniMax H3

Interactive Node Preview — Click the Tabs & ⚙ RUN

Two Ways to Join Shots (pin_mech)

Built-In Finishing (⚙ RUN Page)

1. Realism Boost + Audio Refine (Built for Simple Mode / 1–2 Hop Hero Clips)

2. Refine Pass (hop_refine: full) — Honest Limitations