DreamOmni2: 6 Multi-Image Before and After Edits

DreamOmni2 handles instruction-based image edits with multiple reference images. This post runs 6 before/after edits and shows the raw outputs.

Model

• xiabs/dreamomni2

Test rules

• Each test uses 1 to 3 input images.
• One output per prompt.
• Outputs are published as-is. No manual retouching.

Before and after edits (6 tests)

Edit 1: outfit transfer

Inputs:

Output:

Quick take: clothing transfers look clean when the prompt locks the face and background.

Edit 2: two-person scene composition

Inputs:

Output:

Quick take: multi-subject merges work best when the action stays simple and the environment stays realistic.

Edit 3: pattern transfer onto a car

Inputs:

Output:

Quick take: pattern transfers can hold up well when the prompt also asks for realistic reflections.

Edit 4: cat and dog in one scene

Inputs:

Output:

Quick take: animal edges and fur detail can drift first. Check paws, whiskers, and seat contact shadows.

Edit 5: style transfer between two images

Inputs:

Output:

Quick take: style transfers look best when the prompt locks composition and only moves texture and color.

Edit 6: background replacement on a single image

Input:

Output:

Quick take: background swaps need careful checks on ground contact, shadows, and perspective lines.

Try it

Run DreamOmni2