Krea 2 - Krea AI launches the first basic image generation model trained from scratch.

What is Krea 2?

Krea 2 is Krea AI's first base image generation model trained from scratch, positioned as an "aesthetics-first" creative collaborator. It focuses on visual aesthetic consistency, style transfer, and creative control, rather than simply accurate translation of cue words. Two open-source versions of the model are available.RAW(Undistilled base model, highly malleable, suitable for training LoRA) and Turbo(8-step distillation model, outputs image in about 2 seconds, supports 1K~2K resolution). Krea 2 supports style transfer, mood boards, and four levels of creative parameter adjustment, ranking first among independent lab textural image models in Artificial Analysis's evaluation.

Krea 2's main functions

• Style transferUpload one or more reference images, extract their visual style and apply it to the new image. You can adjust the intensity of the influence of each reference image and mix multiple styles.
• MoodboardsUpload a set of images sharing a creative direction, and Krea 2 can understand the overall tone, texture, atmosphere, and compositional intent, outputting consistent results even with simple prompts.
• CreativityFour-speed controlRaw(Strictly follow the prompts)Low(Approaching prompt words)Medium(Balanced Default)High(The model automatically adds aesthetic depth, composition, and visual richness.)
• Text-to-ImageSupports generating high-quality images from text prompts; the RAW model supports up to 1K resolution, and the Turbo model supports 1K~2K resolution.
• Raw image / Image editingWhen used with the Krea Edit feature, it supports region-level image editing, scene relighting, and outpainting without needing to regenerate the entire image.
• LoRA Training and InferenceTrain LoRA on the RAW base model and run inference on the Turbo distillation model to achieve fast and personalized style customization.
• Krea Nodes WorkflowThe system automatically assembles and processes images by describing the image processing flow in natural language, eliminating the need for manual configuration of complex parameters.

Follow us on WeChat and reply with "open source",join inAI open source project discussion group

How to use Krea 2

• Use directly on the webTo generate a Krea 2 model, visit krea.ai, select the Krea 2 model in the interface, enter the text prompt, upload a style reference image or Moodboards, adjust the Creativity parameters (Raw/Low/Medium/High), and click Generate.
• Local deployment (RAW model)Cloning a GitHub repository krea-ai/krea-2,run uv sync Install dependencies, download RAW model weights and set them. OSS_RAW Environment variables, execution uv run inference.py "提示词" --checkpoint oss_raw --steps 52 --cfg 3.5It supports resolutions up to 1K.
• Local deployment (Turbo model)Download and set the Turbo model weights. OSS_TURBO Environment variables, execution uv run inference.py "提示词" --checkpoint oss_turbo --steps 8 --cfg 0.0 --mu 1.15 --width 2048 --height 2048It can generate images in 8 quick steps and supports 1K~2K resolution.
• Using ComfyUIImport the Krea 2 model into the ComfyUI workflow, load the RAW or Turbo model node, and connect the cue words and sampler nodes for inference.
• API calls: Integrate Krea 2 into your own applications or workflows by calling API services from official partners such as Fal.
• LoRA Training and InferenceTrain LoRA on a RAW model using Hugging Face Diffusers, Ostris AI toolkit, Kohya (musubi tuner), or Fal. After training, load LoRA onto a Turbo model for fast inference.
• Style transfer workflowUpload one or more style reference images, set the influence intensity of each image, and enter prompt words. Krea 2 will automatically blend the reference styles into the new image.
• Moodboards CreationUpload a set of images sharing a common creative direction as a mood board, enter simple prompts, and Krea 2 will generate output with a consistent style based on the overall visual intent.

Krea 2 official website address

• GitHub repositoryhttps://github.com/krea-ai/krea-2

Krea 2's core advantages

• Leading in aesthetic qualityKrea 2 is an "aesthetics-first" design model that ranked first among independent lab text-based image models in Artificial Analysis's evaluation, producing outputs with editor-grade visual quality.
• Open source bi-model architectureAlso provides RAW(Undistilled basic model, extremely flexible) and Turbo(8-step distillation model, outputs graph in about 2 seconds), and the weights of both are completely open source.
• Best practices for RAW training + Turbo inferenceIts unique workflow design balances customization and generation speed by training LoRA on RAW and running inference on Turbo.
• Extremely strong style transfer capabilityIt supports uploading multiple reference images to extract and merge visual styles, and can also understand the collective creative intent of a group of images through Moodboards, achieving batch output with a consistent style.
• Remove AI flavorCompared to other models that produce outputs that are "technically correct but visually bland," Krea 2 is more like a creative collaborator with aesthetic taste, generating results that have less of an "AI feel."
• Creative parameters are flexible and controllableFour levels of Creativity adjustment (Raw/Low/Medium/High) allow you to go from strictly following the prompts to letting the model add aesthetic depth on its own, catering to different creative needs.
• High-resolution fast generationThe Turbo model supports 1K~2K resolution output and only requires 8 sampling steps to achieve high quality.
• Multi-platform ecosystem supportIt officially supports mainstream inference platforms such as ComfyUI, Fal, and SGLang, and provides a complete API and fine-tuning toolchain.
• Regional image editingWhen used with the Krea Edit feature, it supports targeted modifications, relighting, and outward expansion of localized editing without requiring the entire image to be regenerated.

Krea 2's comparison with similar competing products

Application scenarios of Krea 2

• Creative Studios and Concept DesignQuickly explore visual direction, build mood boards, generate concept art and style references, and help the team unify aesthetic direction early in the project.
• Marketing and AdvertisingGenerate consistent editorial/commercial campaign materials, supporting batch output of brand visual style, reducing traditional shooting and post-production costs.
• fashion industryCreate lookbooks and fashion editorial images, precisely control fabric texture, lighting and atmosphere, and overall aesthetic style, and achieve efficient iteration from concept to finished product.
• Architecture and Space DesignIt enables rapid exploration of concept visualization, material research, and spatial composition, helping designers present renderings with various stylistic tendencies during the design phase.
• Game developmentUsed for character design, scene concept, world-building, and prop design, maintaining visual consistency across the entire project through style transfer.
• E-commerce and product displayGenerate high-quality product images in batches with a consistent style, supporting quick switching between different scenes, backgrounds, and styles to enhance the visual appeal of products.