When you attach a reference image in Picture Lab, the AI model treats it as a visual starting point. Instead of generating from the text prompt alone, it builds on the composition, lighting, or style of the image you provide — and applies your prompt as a transformation on top of it.

Attach a reference image

Click References in the prompt bar to open the panel. Select an image from your previous generations, your Favorites, or upload one from your device. The selected image gets a purple border and a number badge showing its selection order. A few models — Flux 2 Klein, for example — use this order when interpreting multiple references. For most models the order has no effect.

References panel with an image selected and prompt ready

Write a transformation prompt

With the reference attached, describe what you want to change or add. The reference handles composition and style — focus your prompt on what should differ.

Generate

Click Generate. The model produces variations that incorporate both the visual reference and your prompt.

The strength of the reference influence depends on the model. If the output drifts too far from the reference, try a more specific prompt that reinforces the key elements you want to keep.

Some models accept more than one reference at a time. If you use multiple, keep them contextually close — pairing very different images can confuse the model and produce inconsistent results.

References work across most generation models in Picture Lab. Flux 1 Kontext Pro uses a dedicated Starter Shot control instead. WAN 2.2 uses Image Prompt.

Edit images with references

Attach a reference image

Write a transformation prompt

Generate