// AI TOOLS · KREA 2

Krea 2

Krea AI's first in-house foundation image model, built for aesthetic range and precise style control.

Last verified · 2026-06-24 · by Moe Ameen

What Krea 2 is

Krea 2 is the first foundation image model Krea AI built from scratch, rather than wrapping someone else's. It is a text-to-image model, and its whole design goal is aesthetic range and creative control instead of literal prompt-following. Krea leans into this on purpose: a short or vague prompt yields a wide spread of high-quality looks rather than one safe default, and you steer the result with style references and moodboards instead of piling adjectives into the prompt. A built-in prompt expander fills in underspecified prompts with richer visual direction.

Krea 2 first launched inside the Krea web product on May 12, 2026, where it generates in roughly 15 seconds or less and is available on a free tier. On June 23, 2026, Krea published a technical report and released open weights in two variants: Krea 2 Raw, an undistilled base checkpoint from mid-training meant for fine-tuning, post-training, and LoRA work, and Krea 2 Turbo, an 8-step distilled checkpoint that generates an image in about two seconds. The intended workflow is "train on Raw, run on Turbo." Weights are on Hugging Face, code is on GitHub, and the model is available through API partners including fal and Replicate.

On capability, Krea 2 ranks in the top tier of the Artificial Analysis text-to-image leaderboard and is among the strongest image models from an independent lab rather than one of the large foundation labs. The open weights ship under a custom license that is free to use but requires companies above roughly 50 seats to take an enterprise agreement, so call it "open weights with a commercial-scale clause" rather than fully unrestricted open source.

Two honest scoping notes. First, Krea 2 is an image model — Krea the platform also offers video and 3D tools, but the Krea 2 model itself does not generate video. Second, at release the model targets resolutions up to 1024px; Krea lists native 2K/4K output, robust editing, and image reference as planned rather than shipped.

What you can make with it

  • Style-consistent images steered by a moodboard or one or more reference images, not just a text prompt
  • A wide spread of distinct looks from a single simple prompt, for fast concepting
  • Near-instant iterations with Krea 2 Turbo (about two seconds per image)
  • Custom fine-tunes — train a LoRA on Krea 2 Raw, then run it on Turbo for fast inference
  • High-aesthetic stills for thumbnails, hooks, ad creative, and reference frames
  • On-brand visual sets where a locked style reference keeps every image in the same look

How Kompozy turns Krea 2 output into content

Krea 2 is exceptional at the look — a clean, on-style, high-aesthetic still that matches your moodboard. What it does not do is turn that still into a finished, scheduled post, and one image is not a content unit. Kompozy is the layer that does. Drop a Krea 2 image into Kompozy and use it as the visual base for a Carousel Post (multi-slide, rendered pixel-exact through HyperFrames), a Quote Graphic, a Photo Post, or a Persona Tweet card — then let the Persona Brief write the captions and copy in your voice, and schedule and publish the set across all nine platforms (Instagram, TikTok, Facebook, YouTube, LinkedIn, X, Pinterest, Threads) plus blog and email from one queue.

The pairing also covers what Krea 2 can't reach at all. From the same idea that produced your Krea 2 still, Kompozy generates the formats a text-to-image model has no path to: persona and avatar video, clipped shorts, blog articles, and newsletters. Krea 2's style references keep the look consistent at the image layer; Kompozy's HyperFrames keep your brand pixel-exact at the post layer and fan one render into a week of cross-platform content instead of a single upload.

  1. Generate an on-style image in Krea 2 using a moodboard or style reference, in Turbo for speed.
  2. Bring the image into Kompozy as the visual base for a Carousel, Quote Graphic, Photo Post, or Persona Tweet.
  3. Let the Persona Brief write the captions and supporting copy in your brand voice; HyperFrames locks the brand styling.
  4. Fan the same concept out into formats Krea 2 can't make — a short, a blog draft, a newsletter, text posts.
  5. Schedule and publish the whole set across TikTok, Reels, Shorts, X, LinkedIn, and the rest from one queue.

Frequently asked questions

What is Krea 2?

Krea 2 is Krea AI's first in-house text-to-image foundation model, built from scratch and tuned for aesthetic range and style control rather than literal prompt-following. It first launched in the Krea web product on May 12, 2026, and Krea released open weights and a technical report on June 23, 2026.

Is Krea 2 an image model or a video model?

Krea 2 is an image model — text-to-image only. The Krea platform separately offers video and 3D tools, but the Krea 2 model itself does not generate video.

What is the difference between Krea 2 Raw and Krea 2 Turbo?

Raw is an undistilled base checkpoint meant for fine-tuning, post-training, and LoRA work. Turbo is an 8-step distilled version that generates an image in about two seconds. Krea's recommended workflow is to train a LoRA on Raw and run inference on Turbo.

Is Krea 2 free or open source?

Krea 2 is available on a free tier in the Krea web product, and the Raw and Turbo weights are released as open weights on Hugging Face under a custom license. The license is free to use but requires companies above roughly 50 seats to take an enterprise agreement, so it is best described as open weights with a commercial-scale clause rather than fully unrestricted open source.

How do I turn Krea 2 images into social posts?

Krea 2 generates the image but does not publish it. Bring the image into Kompozy to build a carousel, quote card, photo post, or tweet card, write the captions in your brand voice via the Persona Brief, and schedule and publish across Instagram, TikTok, LinkedIn, X, Pinterest, and more from one queue.

Related tools

  • HiggsfieldAI video and image platform known for cinematic camera-motion control.
  • MoebiusA 0.2B-parameter image inpainting model that claims to match 10B-scale quality at a fraction of the size.

← All AI tools · Get started →