ComfyUI Node
mimetic-a0 Sampler
Samples like the official tools: Euler over the rectified-flow schedule spaced evenly in log-SNR, classifier-free guidance against the zeroed prompt with adaptive projected guidance, the length plus 6 s of headroom, and the starting noise drawn as the official code does, so a seed gives the same take.
mimetic-a0 Sampler
- model
- positive
- latent
◄seconds8.00►
◄batch_size1►
◄seed0►
◄steps50►
◄cfg6.0►
◄apg_scale1.00►
◄noiseofficial►
Categoryaudio/mimetic_a0
Inputs (9)
| Name | Type | Default | Description |
|---|---|---|---|
| model | MODEL | — | |
| positive | CONDITIONING | CLIPTextEncode with the T5-base text encoder (CLIPLoader type stable_audio) | |
| seconds | FLOAT | 8.000.1–16 | at most 16 s |
| batch_size | INT | 11–16 | takes; one batch of noise, as -n in the command line |
| seed | INT | 00–18446744073709550000 | — |
| steps | INT | 501–500 | 50 is what the model was tested with |
| cfg | FLOAT | 6.01–20 | 6 is what the model was tested with; 1 turns guidance off |
| apg_scale | FLOAT | 1.000–1 | 1: adaptive projected guidance (the official default), 0: plain CFG, in between: a blend |
| noise | COMBO | official | official: drawn on the sampling device like the official code (same seed, same take on the same kind of device). cpu: drawn on the CPU, the same on every machine |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| latent | LATENT | — |