Krea 3: The Next Generation of Image Editing at 8

Krea has unveiled Krea 3, an AI model focused on precise image editing that preserves the original composition while delivering high processing speeds.

New Features

• Regional Editing** — Edit specific areas without regenerating the entire image.
• Precise Mode** — Preserves pixels outside the edited region.
• Multi-reference Editing** — Combines objects, characters, and styles from multiple images (up to 8 references via API).
• Character Consistency** — Maintains character appearance across changes in pose, clothing, and background.
• Product Consistency** — Preserves object geometry, packaging, logos, and labels.
• Text Editing** — Modify text while preserving typography and layout.
• HEX Color Control** — Control colors of individual elements using exact HEX codes.
• Sketch-to-Image** — Transform sketches, drawings, and collages into photorealistic images.
• Native 2K** — Generate 2K resolution images without upscaling.
• Sequential Editing** — Apply sequential changes while retaining previous results.

Technical specifications highlight a claimed editing speed of under 8 seconds. Processing is regional, primarily recalculating only the modified areas.

The fal API offers two options: berry for quality and jam for speed. Settings include reasoning_level (low, medium, high), negative_prompt, seed, and resolution. PNG and JPEG formats are supported.

Base output costs $0.03 for 1K and $0.06 for 2K. Input references and additional modes may incur separate charges.

Speed comparison data from Krea shows the following results:

• Krea 3 — 8 seconds
• Nano Banana Pro — 26.8 seconds
• GPT Image 2.5 — 39.9 seconds

Developers are considering releasing weights and supporting LoRA, but have not made concrete promises yet.