Published: October 18, 2026
Krea 3 is Krea's new image generation and editing model, available on fal.
- Text-to-image generates from prompts with a choice of aspect ratios.
- Image Editing edits and combines multiple reference images from text instructions, with the same aspect ratio options.
The big change from Krea 2, which was text-to-image with style references only, is real editing: restaging products, swapping materials and merging up to 4 references while keeping characters consistent.
Key Features
Editing
- Multi-Reference Compositing: Merge up to 4 images, e.g. a face, an outfit and a backdrop.
- Product Restaging: New scenes and swapped materials, object kept intact.
- Character Consistency: Holds the same character across edits and follow-up shots.
- Targeted Edits: Change one named area and leave camera, framing and layout as they were.
Technical Capabilities
Details |
|
|---|---|
| Quality / Resolution | 1K (default) or 2K |
| Aspect Ratios | 16:9, 9:16, 1:1, 3:4, 4:3 |
| Reference Images | Up to 4 (Image Editing) |
| Prompt Enhancement | Fast (default) or Detailed. Krea refines your prompt before generating; Detailed takes longer |
| Output Format | PNG or JPEG |
Limitations
- Tops out at 2K. No 4K tier, unlike Nano Banana 2, Seedream 5.0 or GPT Image 2.5.
- Editing takes at most 4 references, fewer than the 10 style references Krea 2 accepted.
Prompting Tips
- Give each reference one job, by number: “person from image 1, gown from image 2.”
- Name the exact target (“the wall behind the staircase”) rather than asking for a general redesign.
- Say what must stay: “preserve camera, framing, geometry, lighting and furniture positions.”