ComfyUI Node Runs on cloud
H3 Hires Latent
Encode the upscaled frames of a MiniMax H3 clip into a latent that keeps the clip's sound, for a second pass at the larger size. Sample it from a low sigma, as Pika H3 Flex Sampling's 100, 40, 0, with conditioning at that size.
H3 Hires Latent
- latent
- images
- vae
- latent
CategoryWAS Suite/Latent/Video
Inputs (3)
| Name | Type | Default | Description |
|---|---|---|---|
| latent | LATENT | The first pass's video and audio latent, from its sampler. Its sound is kept unchanged. | |
| images | IMAGE | Every frame of the first pass after upscaling, as 124 for 5 seconds. Width and height are multiples of 16, as 1344x768. | |
| vae | VAE | The MiniMax H3 video VAE, as `minimax_h3_video_vae_int8_convrot`. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| latent | LATENT | Video and audio latent at the frames' size, the first pass's sound in it. |