Bernini-R: Image Edit

Re-light and restage one picture from a written instruction: Bernini-R re-renders the frame with matched light instead of layering a filter over it.

Input

The picture to edit; its framing sets the output size.

What the picture should become.

Advanced settings

Leave empty for a new result each run. Enter a number to repeat one.

Output
Output
Example

Experimente um exemplo

Sobre este fluxo de trabalho

Bernini-R is an editing model built on the Wan stack: one image sets the framing and the scene, and the prompt says what that scene should become — night instead of day, a warm interior instead of a grey studio, a different season. Because it generates the frame rather than compositing over it, the light is rebuilt from the new instruction instead of being pasted on top, so shadows on the ground, reflections and the sky all agree with the change. The conditioning node takes the resized source at length 1, so a still costs one frame of sampling, and a distilled two-stage sampler runs 3 steps. A comparison node shows the source against the result in the same run, which is how you judge whether the relight held the geometry.

Graph: LoadImage -> ResizeImageMaskNode (longer side 1280) -> BerniniConditioning (length 1) -> two-stage SamplerCustom (high/low noise, 3 steps, res_multistep) -> VAEDecode -> SaveImage (png), plus ImageCompare of the source against the result.