ComfyUI Node
H3 Auto Prompt Chain v2
Optimized one-click H3 prompt-list chain: dynamic per-shot step/context budgets, direct AV latent continuation, resumable latent saves, timing profiler and one final memory-bounded decode.
H3 Auto Prompt Chain v2
- model
- clip
- video_vae
- audio_vae
- sampler
- first_frame
- video_path
- run_info
- profile_path
◄global_promptGLOBAL CONTINUITY:
Preserve established character identity, facial features, body proportions, wardrobe, scene geometry, object state, lighting logic and cinematic style across shots.
Natural temporal motion and coherent native audio.
non_diegetic_music: N/A►
◄shot_prompts[BALANCED] SHOT 01: A woman walks slowly through a quiet apartment toward the window.
---
[FAST] SHOT 02: She reaches the window and looks down at the street.
---
[MOTION] SHOT 03: She suddenly turns and walks quickly toward the table.►
◄width608►
◄height352►
◄default_duration_seconds8.0►
◄default_steps6►
◄schedulersimple►
◄default_context_frames22►
◄manual_tail_frames7►
◄base_seed123456789►
◄start_clip1►
◄latent_prefixh3_autochain/clip►
◄output_prefixvideo/H3_AUTO_CHAIN_FINAL►
◄crf18►
CategoryMiniMax H3/automation
Inputs (20)
| Name | Type | Default | Description |
|---|---|---|---|
| model | MODEL | — | |
| clip | CLIP | — | |
| video_vae | VAE | — | |
| audio_vae | VAE | — | |
| sampler | SAMPLER | — | |
| global_prompt | STRING | GLOBAL CONTINUITY: Preserve established character identity, facial features, body proportions, wardrobe, scene geometry, object state, lighting logic and cinematic style across shots. Natural temporal motion and coherent native audio. non_diegetic_music: N/A | — |
| shot_prompts | STRING | [BALANCED] SHOT 01: A woman walks slowly through a quiet apartment toward the window. --- [FAST] SHOT 02: She reaches the window and looks down at the street. --- [MOTION] SHOT 03: She suddenly turns and walks quickly toward the table. | Separate multiline shots with ---. Optional leading tags: [FAST], [BALANCED], [QUALITY], [dur=10.1], [steps=4], [ctx=5], [tail=7]. |
| width | INT | 60832–4096 | — |
| height | INT | 35232–4096 | — |
| default_duration_seconds | FLOAT | 8.05–15 | — |
| default_steps | INT | 61–40 | — |
| scheduler | COMBO | simple | 9 options: simple, sgm_uniform, karras, exponential, ddim_uniform, beta, +3 |
| default_context_frames | COMBO | 22 | 3 options: 5, 22, 39 |
| manual_tail_frames | INT | 70–68 | Fast no-decode handover tail. 7 mirrors the suite's balanced no-lock fallback request before phase alignment. |
| base_seed | INT | 1234567890–18446744073709550000 | — |
| start_clip | INT | 11–80 | 1=fresh run. If clip N failed, keep saved latents and resume with N. |
| latent_prefix | STRING | h3_autochain/clip | — |
| output_prefix | STRING | video/H3_AUTO_CHAIN_FINAL | — |
| crf | INT | 180–51 | — |
| first_frameopt | IMAGE | Optional first-frame anchor for clip 1. Leave disconnected for pure T2V. |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| video_path | STRING | — |
| run_info | STRING | — |
| profile_path | STRING | — |