ComfyUI Node
FL MiniMax H3 Prompt Timeline
Creates MiniMax H3 video/audio latents with both strict latent-time prompt scheduling and a fast semantic timeline conditioning.
FL MiniMax H3 Prompt Timeline
- clip
- vae
- audio_vae
- prompt_schedule
- prompt_envelopes
- ref_images
- ref_videos
- ref_video_audios
- ref_audios
- scheduled
- latent
- semantic
- timeline
◄global_prompt—►
◄width1344►
◄height768►
◄length124►
◄time_unitseconds►
◄bpm120.0►
◄transition_modecosine►
◄transition_frames4►
◄affect_audiovideo only►
◄duration_policystrict►
◄ref_image_sizematch►
◄timeline►
CategoryFL/MiniMax H3/Prompting
Inputs (21)
| Name | Type | Default | Description |
|---|---|---|---|
| clip | CLIP | MiniMax H3 text encoder used for the global and scheduled prompts. | |
| vae | VAE | MiniMax H3 video VAE used for reference images and reference videos. | |
| audio_vae | VAE | MiniMax H3 audio VAE used for audio references and video soundtracks. | |
| global_prompt | STRING | Persistent subject, style, and scene context. This is the default conditioning under scheduled masks and is prepended to every scheduled prompt. | |
| width | INT | 134432–16384 | Width of the native H3 video latent in output pixels. |
| height | INT | 76832–16384 | Height of the native H3 video latent in output pixels. |
| length | INT | 1245–3600 | Requested frame count at H3's fixed 24 fps. H3 snaps it to the 17k+5 frame grid. |
| time_unit | COMBO | seconds | Legacy unit used only by manual timeline headers. Connected prompt schedules own their time unit; H3 video always runs at 24 fps. |
| bpm | FLOAT | 120.01–999 | Legacy constant BPM used only for manual timelines in beats mode. Connected prompt schedules carry exact detected timing. |
| transition_mode | COMBO | cosine | Transition curve between contiguous sections in the manual timeline. |
| transition_frames | INT | 40–96 | Manual timeline transition width in 24 fps video frames. |
| affect_audio | COMBO | video only | Choose whether scheduled text masks affect only H3 video tokens or both generated video and audio tokens. |
| duration_policy | COMBO | strict | strict rejects ranges past the H3 duration; clamp trims them; fit scales the full schedule to the aligned H3 duration. |
| ref_image_size | COMBO | match | match limits reference image area to the output canvas; max uses H3's maximum reference short edge. |
| timelineopt | STRING | Legacy manual ordered time ranges followed by prompt text. Used only when prompt_schedule is not connected. | |
| prompt_scheduleopt | FL_PROMPT_SCHEDULE | Optional exact-time schedule, such as the output of FL Audio Beat Prompt Schedule. When connected, it overrides the manual timeline. | |
| prompt_envelopesopt | COMFY_AUTOGROW_V3 | Optional audio-reactive prompt envelopes. Each unique prompt adds one conditioning evaluation per sampling step. | |
| ref_imagesopt | COMFY_AUTOGROW_V3 | Optional H3 reference images. Add sockets as needed. | |
| ref_videosopt | COMFY_AUTOGROW_V3 | Optional H3 reference videos. Add sockets as needed. | |
| ref_video_audiosopt | COMFY_AUTOGROW_V3 | Optional soundtracks paired by index with reference videos. | |
| ref_audiosopt | COMFY_AUTOGROW_V3 | Optional standalone H3 audio references. Add sockets as needed. |
Outputs (4)
| Name | Type | Description |
|---|---|---|
| scheduled | CONDITIONING | Strict latent-time conditioning for the first H3 sampling pass. |
| latent | LATENT | Native nested MiniMax H3 video/audio latent. |
| semantic | CONDITIONING | Single semantic timeline conditioning for faster low-denoise refinement passes. |
| timeline | FL_H3_TIMELINE | Reusable encoded H3 timeline for applying the same schedule after spatial resizing. |