ComfyUI Node Runs on cloud
H3 Stream Sampler
Sample MiniMax H3's picture and sound in one pass with different step rules, by default the picture as lcm and the sound as Euler ancestral. The sound steps on its own sigmas from the model's audio shift.
H3 Stream Sampler
- sampler
◄videolcm►
◄audioeuler_ancestral►
◄eta1.00►
CategoryWAS Suite/Sampling
Inputs (3)
| Name | Type | Default | Description |
|---|---|---|---|
| video | COMBO | lcm | Picture step: `lcm` = clean prediction re-noised to the next sigma, as Pika H3 Flex; `euler` = no fresh noise; `euler_ancestral` = fresh noise by eta. |
| audio | COMBO | euler_ancestral | Sound step, on the sound's own sigmas: `euler_ancestral` = fresh noise by eta; `lcm` = re-noised clean prediction; `euler` = no fresh noise. |
| eta | FLOAT | 1.000–1 | Fresh noise in an `euler_ancestral` step: `1.0` = full; `0` = euler. Other rules ignore it. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| sampler | SAMPLER | For Sampler Custom Advanced on MiniMax H3, with picture sigmas such as Pika H3 Flex Sampling's. `lcm` on both = core's lcm. |