Skip to content

Image to image

A photo from the gallery, re-imagined at the strength you choose.

Nodes: Prompt and Image → Image to image → Output.

Use it to restyle a photo (a painting of it, an anime version), or to take a rough picture and let the model redraw it with more detail.

FLUX.2 and Qwen Image use Image edit instead

These two are edit models: give them a photo and they re-render it to follow an instruction. That is its own flow. See Image edit.

Steps

  1. Flows → Image to image.
  2. Tap the image node and choose a photo. It appears on the node straight away.
  3. Tap the prompt node and describe the result you want. Describe the whole picture, not just the change.
  4. Tap the Image to image node and set Denoise (below).
  5. Run.

Draw on it first

Under the picture on the image node, Draw opens a drawing window over the photo: the brush, the eraser and the eyedropper as icons, undo, redo and clear, colour swatches, and one slider whose dropdown picks what it sets (Size, Opacity, Softness or Hue). Two fingers zoom and pan for detail work; one finger always paints. The eyedropper takes the colour where your finger lifts, and the rainbow + swatch opens a picker for any colour. Paint a rough shape or a blob of colour where you want something, pull the window down, and run: the drawing saves itself as you go. Image to image follows your colours and shapes (a denoise around 0.6–0.75 turns a doodle into something real). The photo itself is kept: the drawing is a layer on it, so you can reopen it, erase it, or Clear drawing.

No photo? An empty image node offers two halves: Add image to pick one, or Blank page to start on a white page and draw.

Denoise: how much changes

Denoise runs from 0 to 1:

denoise result
0.2 – 0.4 Close to the photo: colours and details shift, the composition stays
0.5 – 0.7 A real reinterpretation that keeps the layout (0.65 is the default)
0.8 – 1.0 Mostly new; the photo only guides the overall shapes

Framing the photo

The generator draws at a fixed size and shape, so the photo is framed to that shape first. Tap the picture on the generator node (or Crop in its sheet) to open the framing view: drag the photo to move it and pinch to zoom; the frame is exactly what will be sent.

You can also zoom out past the photo's edges (up to twice its area), which makes the subject smaller in the frame. The part outside the photo shows as a checkerboard: the model sees it filled with the photo's own edges, mirrored and blurred, but it is cut away from the result, so only the photo's own area comes back. The photo is never stretched.

The framing view for image to image

Zoomed out a little: the checkerboard strip is not part of the result.