ComfyUI Node
kubakub keyframe clips (h3)
Turns the clips you set on the director's timeline into generated video with sound and puts them into its sequence (MiniMax H3, run locally). Keyframe clips (the fl2va, first and last frame, model): the frames at the clip's start and end become H3's first and last frame, H3 animates the way between them. Reference clips (the ref2va, reference, model): director layers, the music of that stretch and / or the stretch itself as references. H3's sound is mixed into the timeline sound.
kubakub keyframe clips (h3)
- model
- clip
- vae
- audio_vae
- frames
- audio
- model_ref
- layer_images
- enhance_clip
- frames
- prompts
- audio
- report
- fps
◄fps25.00►
◄document►
◄steps4►
◄steps_ref20►
◄megapixels0.85►
◄seed42►
◄h3_audio0.60►
◄crossfade6►
◄enhancetrue►
Categorykubakub/2d/motion
Inputs (18)
| Name | Type | Default | Description |
|---|---|---|---|
| model | MODEL | H3 fl2va (keyframe clips), e.g. + the 4-step flashgen LoRA. | |
| clip | CLIP | qwen3vl 32b MiniMax H3 encoder (CLIPLoader type minimax). | |
| vae | VAE | H3 video VAE. | |
| audio_vae | VAE | H3 audio VAE. | |
| frames | IMAGE | The frames output of kubakub director sequence. | |
| fps | FLOAT | 25.001–120 | The director's fps output. |
| document | STRING | The director's document output (its timeline clips). | |
| steps | INT | 41–100 | Keyframe clips (4 with the flashgen LoRA). |
| steps_ref | INT | 201–100 | Reference clips (20 without a turbo LoRA, 4 with). |
| megapixels | FLOAT | 0.850.1–2 | H3 work size (the aspect of the sequence); scaled to the sequence size after. |
| seed | INT | 420–18446744073709550000 | Random seed for the clips (each clip adds its number to it); another seed gives other motion. |
| h3_audio | FLOAT | 0.600–2 | Level of H3's own sound in the mix (0 = keep only the timeline sound). |
| crossfade | INT | 60–48 | Frames blended into the director's own frames at each clip end (H3 does not always land exactly on the end frame). At most 12 are used. |
| audioopt | AUDIO | The director's audio output (timeline sound). | |
| model_refopt | MODEL | H3 ref2va for reference clips (+ a turbo LoRA -> fewer steps). | |
| layer_imagesopt | IMAGE | The images connected to the director's 'layers' input (for references to those layers). | |
| enhance_clipopt | CLIP | A text encoder that can write (the Gemma 4 LTX encoder: CLIPLoader type ltxv) - it expands each clip's prompt into H3's format, looking at the clip's first frame. | |
| enhanceopt | BOOLEAN | true | Use enhance_clip when connected. |
Outputs (5)
| Name | Type | Description |
|---|---|---|
| frames | IMAGE | The sequence with the clips in. |
| prompts | STRING | The prompts H3 got, per clip (copy one into a clip with 'raw' on to tune it). |
| audio | AUDIO | Timeline sound + H3 sound. |
| report | STRING | Per clip: mode, time span, H3 size, frames, steps and time (or 'from the cache'). |
| fps | FLOAT | The frame rate (frames per second), passed through for Create Video. |