ComfyUI Node
Bernini Full Reference to Video
Full Bernini reference-image-to-video using Bernini-Diffusers only.
Bernini Full Reference to Video
- reference_image
- video
◄promptAnimate the reference image into a short smooth video.►
◄negative_promptlow quality, blurry, distorted, text, watermark►
◄num_frames9►
◄fps16►
◄width512►
◄height512►
◄steps4►
◄seed5678►
◄quality_presetstandard►
◄accelerationwan2.2_lightx2v_4step►
◄memory_modekeep_loaded_batch►
CategoryRunningHub/Bernini
Inputs (12)
| Name | Type | Default | Description |
|---|---|---|---|
| reference_image | IMAGE | One or more reference images. Batched IMAGE inputs are passed as multiple references. | |
| prompt | STRING | Animate the reference image into a short smooth video. | Text prompt or edit instruction. |
| negative_prompt | STRING | low quality, blurry, distorted, text, watermark | Negative prompt. |
| num_frames | INT | 95–129 | Number of output frames. Higher values increase VRAM use and runtime. |
| fps | INT | 161–60 | Frame rate written into the returned video. |
| width | INT | 51264–2048 | Output width in pixels. Larger values increase VRAM use and runtime. |
| height | INT | 51264–2048 | Output height in pixels. Larger values increase VRAM use and runtime. |
| steps | INT | 41–100 | Diffusion steps. |
| seed | INT | 56780–4294967295 | Random seed. Values are clamped to NumPy's valid 0 to 2**32-1 range before inference. |
| quality_preset | COMBO | standard | standard balances speed and quality; quality spends more planning time for better output. |
| acceleration | COMBO | wan2.2_lightx2v_4step | Auto-load the Wan2.2 T2V LightX2V 4-step LoRA pair from ComfyUI/models/loras when available. |
| memory_mode | COMBO | keep_loaded_batch | keep_loaded_batch keeps the Bernini pipeline cached for faster consecutive jobs; low_vram_unload frees memory after each run. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| video | VIDEO | — |