How multi-reference models combine images
Older tools stitched layers together. Multi-reference models instead read every image you upload and draw a new picture that uses what you point to — the subject of one, the setting of another, the style of a third — and relight it so it looks like one photo.
A prompt pattern that works
Most good combine prompts follow this order:
- What goes where: “the mug from image 1 on the windowsill in image 2”.
- What must stay: “keep her face, hair and pose”.
- The light and mood of the final scene: “same blue-hour light”.
- Any small detail that matters, named explicitly.



























