Krea 2
Krea 2 is an AI tool in the Image & Design category. Useful for open weights, image generation, comfyui, lora, text to image.
Open weights image model from Krea, trained from scratch, shipped as Raw for training LoRAs and Turbo for 8-step generation.
Modelo de imagem de pesos abertos da Krea, treinado do zero, com o Raw pra treinar LoRA e o Turbo pra gerar em 8 passos.
AITag Review
Krea 2 is an image model trained from scratch, not a FLUX fine-tune: a single-stream transformer reads text and image in one sequence, and a frozen Qwen3-VL reads the prompt as a description of color, shape, texture and count, so quoted text comes out written. The open release has two checkpoints: Raw, undistilled, for LoRA training and research, and Turbo, distilled to 8 steps with CFG off, for generation from 1K to 2K. ComfyUI has an official repack, and the nvfp4 Turbo fits in 7.7 GB, so a 16 GB RTX 50 card makes a one megapixel image in about 16 to 21 seconds. Commercial use is allowed under the Krea 2 Community License for companies under one million dollars of annual revenue; above that, Krea offers an enterprise license.
Features
- Category: Image & Design
- Tags:open weightsimage generationcomfyuiloratext to image
- Website:Visit
- Curated: No
Related tools
Social AI platform offering a variety of LoRAs and advanced models for image generation. Allows experimentation with different models and visual styles.
Advanced image fusion tool developed by Google. Uses artificial intelligence to intelligently combine multiple images, allowing the creation of complex visual compositions. Employs techniques like image analysis, segmentation, and synthesis to ensure visual coherence and realism, enabling users to merge elements from different sources into cohesive new visuals.
OpenAI AI model for generating images from text. Produces highly accurate, detailed visuals and is integrated into ChatGPT Plus.
AI-powered platform focused on image generation and visual enhancement. Provides multiple tools to create and edit visuals, including upscaling, variation generation, and custom visual style options.