ComfyUI Node
FL MiniMax H3 Beat Shot Planner
Turns an FL beat prompt schedule into MiniMax H3 renders. Prompt sections grouped in the audio sequencer share one render; ungrouped sections remain independent.
FL MiniMax H3 Beat Shot Planner
- clip
- vae
- audio_vae
- prompt_schedule
- timeline_audio
- prompt_envelopes
- ref_images
- ref_videos
- ref_video_audios
- ref_audios
- shot_plan
- scheduled
- latent
- semantic
◄length124►
◄global_prompt—►
◄width1344►
◄height768►
◄affect_audiovideo only►
◄ref_image_sizematch►
◄visual_condition_fidelity1.00►
◄visual_reference_modefull►
◄reference_strength1.00►
CategoryFL/MiniMax H3/Prompting
Inputs (19)
| Name | Type | Default | Description |
|---|---|---|---|
| clip | CLIP | MiniMax H3 text encoder used for each planned render. | |
| vae | VAE | MiniMax H3 video VAE used to encode image and video references. | |
| audio_vae | VAE | MiniMax H3 audio VAE used to encode each shot's matching audio slice. | |
| length | INT | 1245–3600 | Requested frame count for a single render, used only when prompt_schedule is not connected. H3 snaps it to the 17k+5 frame grid. |
| global_prompt | STRING | Persistent character, identity, visual style, and production context. | |
| width | INT | 134432–16384 | Width of every planned render. |
| height | INT | 76832–16384 | Height of every planned render. |
| affect_audio | COMBO | video only | Choose whether scheduled prompt masks also affect H3 audio tokens. |
| ref_image_size | COMBO | match | Reference image sizing, matching the standard MiniMax H3 reference node. |
| visual_condition_fidelity | FLOAT | 1.000–1 | Noise fidelity for H3 visual latent references and motion context. Lower values replace more latent detail with seeded noise. This does not scale Qwen-VL. |
| visual_reference_mode | COMBO | full | Select which visual reference paths receive connected images and videos. Reference audio and video soundtracks remain active in every mode. |
| reference_strength | FLOAT | 1.000–1 | Blend from a prediction without connected visual references at 0 to the selected mode at 1. Audio and prior-shot motion context remain in both. Beat sampler cfg multiplies this value; effective blends other than 0 or 1 require both predictions while sampling. |
| prompt_scheduleopt | FL_PROMPT_SCHEDULE | Exact 24 fps shot ranges from FL Audio Beat Prompt Schedule. Optional; when disconnected the node plans one full-length render from the length input. | |
| timeline_audioopt | AUDIO | Frame-aligned cropped audio from FL Audio Beat Prompt Schedule. Each shot receives only its matching slice. Optional; when disconnected each planned render has no shot-local audio reference. | |
| prompt_envelopesopt | FL_PROMPT_ENVELOPE_SET | Sequence-wide reactive prompt envelope set from FL Audio Beat Prompt Sequencer, sliced and rebased for every planned render. | |
| ref_imagesopt | COMFY_AUTOGROW_V3 | Character or scene images applied to every planned render. | |
| ref_videosopt | COMFY_AUTOGROW_V3 | Reference videos applied to every planned render. | |
| ref_video_audiosopt | COMFY_AUTOGROW_V3 | Soundtracks paired by index with reference videos. | |
| ref_audiosopt | COMFY_AUTOGROW_V3 | Additional global audio references applied after the shot-local timeline audio. |
Outputs (4)
| Name | Type | Description |
|---|---|---|
| shot_plan | FL_H3_SHOT_PLAN | Planned nested H3 latents and conditioning for sequential sampling. |
| scheduled | CONDITIONING | Positive conditioning. Meaningful for the single render produced when prompt_schedule is not connected; with a schedule it reflects the first render. |
| latent | LATENT | Native nested MiniMax H3 video/audio latent. Meaningful for the single render produced when prompt_schedule is not connected; with a schedule it reflects the first render. |
| semantic | CONDITIONING | Single global conditioning. Used for the manual single render; with a schedule it reflects the first render. |