Krea 3
Krea's image-editing model: 2K edits in under 8 seconds, region and bounding-box control, up to six references, and LoRA fine-tuning.
Overview
Krea opened the public beta of Krea 3 on October 15, 2026, an image-editing model that changes a photo or reference images from a written instruction rather than generating from scratch. Edits can be limited to a selected region or a bounding box, run over multiple turns, and use up to six reference images, including style references. Krea claims 2K edits finish in under 8 seconds end to end and says Krea 3 is the first editing model that can be fine-tuned with LoRAs on a specific product, person, or style. Quality parity with Nano Banana Pro and GPT Image 2.5 is Krea's own claim; independent tests were thin at launch.
This is a sourced directory entry. Features and capabilities are reported in the linked sources; this entry does not document independent Aymorphic testing.
Key features
- Instruction-based image editing from a photo or reference images
- Regional edits and bounding boxes that leave the rest of the frame unchanged
- Up to six reference images per edit, including style references
- Multi-turn editing sessions
- LoRA fine-tuning on a product, person, or style (early-access approval reported for training)
- Native 2K output with support up to 4K; 2K edits in under 8 seconds (Krea-reported)
- Available on krea.ai, the Krea API, and partners including fal, Together AI, OpenRouter, Replicate, and Baseten
Use cases
- Product photography retouching and background changes
- Keeping a person or product consistent across a batch of edited images
- Style-matched creative edits using reference images
Pricing
Usage-based: $0.03 per 1K image and $0.06 per 2K image; enterprise access by request. Beta, so terms may change.
Pricing reflects the source record and may change. Confirm current terms with the provider.
Dates and sources
- Released
- 15 Oct 2026
- Record updated
- 18 Oct 2026
- Record last checked
- 18 Oct 2026