ComfyUI Node
BSAI Sol-H3 Latent Upscale Align (Two-Pass)
Two-pass Self-Lift: latent upscale (no VAE) + pixel 32-align + CONST re-noise; output feeds SamplerCustomAdvanced + DisableNoise for second pass
BSAI Sol-H3 Latent Upscale Align (Two-Pass)
- samples
- model
- noise
- sigmas
- latent
◄scale2.0►
◄align_to32►
◄methodbilinear►
◄upscaler_model(bilinear, no model)►
◄add_noisetrue►
◄audio_denoise0.00►
CategoryBSAI/Sol-H3
Inputs (10)
| Name | Type | Default | Description |
|---|---|---|---|
| samples | LATENT | — | |
| scale | FLOAT | 2.01–4 | latent upscale ratio; 1.5-2.0 recommended for two-pass (target resolution = first pass x ratio) |
| align_to | INT | 328–256 | pixel resolution alignment step (H3 official: 32px); avoids edge color bands |
| method | COMBO | bilinear | 5 options: nearest-exact, area, bilinear, bicubic, bislerp |
| upscaler_model | COMBO | (bilinear, no model) | choose a 3D upscaler model for neural semantic upscale (richer detail than bilinear); pick bilinear for plain interpolation |
| add_noise | BOOLEAN | true | True = CONST re-noise to sigmas[0] for second pass; False = upscale-align only |
| audio_denoise | FLOAT | 0.000–1 | 0 = lock first-pass audio for second pass (default); >=0.5 = re-noise and redraw audio together |
| modelopt | MODEL | — | |
| noiseopt | NOISE | — | |
| sigmasopt | SIGMAS | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| latent | LATENT | — |