ComfyUI Node
Bernini-R Conditioning
Creates latent and attaches in-context visual references for Bernini-R. Encodes source_video, reference_video, and reference_images to context_latents with chunked encoding for VRAM efficiency. Supports batch_size > 1 for parallel generation.
Bernini-R Conditioning
- positive
- negative
- vae
- source_video
- reference_video
- reference_images
- mask
- positive
- negative
- latent
◄width832►
◄height480►
◄length81►
◄batch_size1►
◄ref_max_size848►
◄chunk_frames16►
◄chunk_overlap4►
◄mask_modeanneal►
CategoryBernini-R/Conditioning
Inputs (15)
| Name | Type | Default | Description |
|---|---|---|---|
| positive | CONDITIONING | — | |
| negative | CONDITIONING | — | |
| vae | VAE | — | |
| width | INT | 83264–4096 | — |
| height | INT | 48064–4096 | — |
| length | INT | 811–10000 | — |
| batch_size | INT | 11–64 | — |
| ref_max_size | INT | 84864–4096 | — |
| chunk_frames | INT | 161–256 | Max pixel frames per VAE encoding chunk. Snapped to nearest multiple of 4 internally. |
| chunk_overlap | INT | 44–32 | Pixel frames of overlap between chunks. Prevents temporal boundary artifacts from VAE 3D convolutions. |
| source_videoopt | IMAGE | — | |
| reference_videoopt | IMAGE | — | |
| reference_imagesopt | IMAGE | — | |
| maskopt | MASK | Mask of the region to edit. White (1) = regenerate, black (0) = keep source. From a segmenter like SAM2; connect directly without inverting. Connecting enables differential diffusion. | |
| mask_modeopt | COMBO | anneal | anneal = soft background anchoring with natural boundaries (default); freeze = pixel-level freeze, background stays completely still |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| positive | CONDITIONING | — |
| negative | CONDITIONING | — |
| latent | LATENT | — |