ComfyUI Node
H3 Encode
MiniMax H3 conditioning and its AV latent, from a prompt plus an optional H3 Inputs list. Covers all three of H3's conditioning tasks — t2va (prompt only), fl2va (first/last keyframes) and ref2va (references) — replacing both MiniMax H3 Image to Video and MiniMax H3 Reference to Video: the task follows from the roles in the list, so switching between keyframes and references is a row setting rather than a rewire.
H3 Encode
- clip
- vae
- audio_vae
- h3_inputs
- positive
- LATENT
- report
◄prompt—►
◄modeauto►
◄keyframe_instructionauto►
CategoryNynxz/H3
Inputs (7)
| Name | Type | Default | Description |
|---|---|---|---|
| clip | CLIP | H3's Qwen3-VL-32B text encoder. | |
| vae | VAE | H3's video VAE. | |
| prompt | STRING | Refer to references by the tag the H3 Image / Video / Audio node reports for them — <Picture 1>, <Video 1>, <Audio 1>. They are numbered in wire order, counting each kind separately. | |
| mode | COMBO | auto | Which of H3's three tasks to run. t2va is prompt only; fl2va anchors a first and/or last frame in the clip; ref2va uses references with no fixed position. auto picks fl2va if anything is wired as a first or last frame, ref2va if anything else is wired, t2va otherwise. Force one when both kinds are wired — the model cannot take both at once. Forcing fl2va with nothing marked uses the first two pictures as the first and last frame. |
| keyframe_instruction | COMBO | auto | H3 expects a keyframe prompt to OPEN with a line saying where each picture lands in the clip — see VIDEO_PROMPT_WRITING_GUIDE_base_en section 2.1. auto writes it for you, using the snapped duration (ask for 130 frames and the correct mark is 5.88s, not 5.42s). Skipped if your prompt already has one. Turn off for a multi-shot prompt, which needs its own shot index. |
| audio_vaeopt | VAE | H3's audio VAE. Only needed when a reference carries audio — a dropped video's soundtrack, or an audio row. | |
| h3_inputsopt | NYNXZH3_INPUT | From H3 Canvas and/or a chain of H3 Image / Video / Audio nodes — the assets and the canvas they are framed for. Unwired, this is plain text-to-video at H3's default 1344x768, 124 frames. |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| positive | CONDITIONING | — |
| LATENT | LATENT | — |
| report | STRING | Which task ran, the canvas it ran at, and anything set aside and why. |