Krea opened a beta for Krea 3, an editing-first model that changes a selected region of an image in under eight seconds and can be fine-tuned with LoRAs to keep a product or character consistent.
2026-10-15 · by Moe Ameen
Krea AI opened a beta for Krea 3 on October 15, 2026. Unlike Krea 2, which is a text-to-image model that generates a picture from a prompt, Krea 3 is an editing model: you give it a photo or a set of reference images and a written instruction, and it changes the image. The headline feature is regional editing — you select an area to change and leave the rest of the frame untouched, and bounding boxes let you place new elements in a specific spot, so a local change like recoloring an object or swapping a background element doesn't regenerate the person or the rest of the scene.
Krea says Krea 3 returns high-quality 2K edits in under eight seconds end to end, supports up to 4K for print and campaign assets, and accepts up to six reference images per edit, including style references. (On fal's edit endpoint the reference limit is listed as four per request.) Pricing is usage-based: Krea lists $0.03 per image at 1K and $0.06 per image at 2K, with no subscription on the pay-as-you-go API partners. Krea also claims Krea 3 is the first editing model that can be fine-tuned with LoRAs on a specific product, person, or style, so a packshot or recurring character stays consistent across edits.
Krea positions the model's quality as comparable to Nano Banana Pro and GPT Image 2.5, citing its own benchmark figures; independent testing is still thin at launch, so treat the quality and speed numbers as company-reported until third parties confirm them. Krea 3 is reachable on krea.ai and the Krea API, and through partners including fal, Together AI, OpenRouter, Replicate, Baseten, Hedra, and Nous Research.
Krea 3 solves the "fix this one thing" problem — recolor a jacket, swap a background, drop in a product — fast and cheaply. What it does not do is turn those edited stills into posts. That is the handoff to Kompozy. Edit your hero image or product shot in Krea 3, then bring it into Kompozy as the visual base for a brand-exact Carousel through HyperFrames, a Quote Graphic, a Photo Post, or a Persona Tweet card. Kompozy writes the captions in your voice from the Persona Brief, generates the formats Krea 3 can't touch — persona and avatar video, clips, blogs, newsletters — and schedules and publishes the finished set across eight social platforms plus blog and email from one queue.
The high-volume editing angle pairs especially well with Kompozy's batch pipeline. Run a catalog of product edits through Krea 3, feed the images into Kompozy, and let autopilot turn them into a week of carousels, photo posts, and captioned shorts routed to each platform on a schedule — so the cheap edits become finished, on-brand content instead of a folder of PNGs.
Krea 3 is Krea AI's image-editing model, launched in beta on October 15, 2026. You give it a photo or reference images and a written instruction and it edits the image, with regional control over which area changes. It is distinct from Krea 2, which generates images from a text prompt.
Krea lists $0.03 per image at 1K resolution and $0.06 per image at 2K, pay-as-you-go on API partners. Krea says 2K edits return in under eight seconds end to end; those figures are company-reported at launch.
Krea 2 is a text-to-image model that generates a new image from a prompt. Krea 3 is an editing model that modifies an existing image or set of references, with regional editing, up to six reference images, and LoRA fine-tuning for product and character consistency.
Krea 3 edits the image but does not publish it. Bring the edited image into Kompozy to build a carousel, quote card, or photo post, write captions in your brand voice, and schedule and publish across eight social platforms plus blog and email from one queue.