ComfyUI Node
H3 Clip Refiner /noEmbryo
Inline texture-ratchet corrector for H3 Motion Context clips. Place between SamplerCustomAdvanced and H3 Motion Context Save Latent. Fixes the texture ratchet: high-band noise/grain that increases at each join. target_ratio=0: measures band_ratio (high-band std / total std), passes latent through. target_ratio>0: applies match_band to rescale high-band to target ratio. strength=0: measure only. strength=1: full correction. >1: over-correction. Run first clip with target_ratio=0 to measure its band_ratio, then set target_ratio to that value for all subsequent clips. To make it LESS grainy, set target_ratio LOWER than measured band_ratio. Use strength>1.0 for over-correction if needed.
H3 Clip Refiner /noEmbryo
- latent
- reference_latent
- latent
- report
◄target_ratio0.0000►
◄strength1.00►
CategorynoEmbryo/MiniMax H3
Inputs (4)
| Name | Type | Default | Description |
|---|---|---|---|
| latent | LATENT | The generated AV latent from your H3 sampler (SamplerCustomAdvanced output). | |
| target_ratio | FLOAT | 0.00000–1 | 0.0 = measure only (report band_ratio, pass latent through). >0.0 = target band_ratio to match. Set to the first clip's band_ratio (e.g., 0.3643) to fix the texture ratchet. To make it LESS grainy, set target_ratio LOWER than the measured band_ratio (e.g., measured 0.3917 -> set 0.3800). |
| strength | FLOAT | 1.000–3 | Correction strength. 0.0 = measure only (pass through). 1.0 = full correction to target_ratio. >1.0 = over-correction (stronger grain reduction). Use >1.0 if target_ratio alone doesn't reduce grain enough. |
| reference_latentopt | LATENT | Optional: reference latent (e.g., first clip) to measure target_ratio from automatically. If connected, target_ratio is ignored and measured from this latent. |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| latent | LATENT | — |
| report | STRING | — |