CtrlK
Krea AI (stylized as KREA) is a San Francisco-based generative AI startup that builds a unified creative platform for real-time AI image generation, video generation, image upscaling, and 3D object creation. Founded in March 2022 by Victor Perez (CEO) and Diego Rodriguez (CTO), Krea has positioned itself as the market leader in real-time image generation for creative professionals, producing photorealistic images from simple inputs in under 50 milliseconds. By mid-2025, the platform had attracted over 20 million users across 191 countries. In April 2025, Krea raised $47 million in Series B funding led by Bain Capital Ventures, bringing its total funding to $83 million and its valuation to approximately $500 million.
Krea integrates multiple third-party AI models (including Flux, Veo 3, Kling, Wan 2.1, and Hailuo) into a single platform, allowing designers, photographers, and artists to access different generation engines without switching between tools. The company has also developed its own proprietary model, Krea 1, in collaboration with Black Forest Labs.
Victor Perez and Diego Rodriguez met as undergraduate students in Barcelona in the early 2010s while studying Audiovisual Systems Engineering. Both had creative backgrounds: Perez came from graffiti art and graphic design, while Rodriguez had experience in digital art and 3D architectural visualization. Rodriguez introduced Perez to coding and generative adversarial networks, sparking Perez's interest in using artificial intelligence for creative work.
After graduating, both received prestigious full scholarships from the King of Spain to attend Cornell University for graduate studies. Rodriguez began his program in New York, but when Perez arrived at Cornell a semester later, he stayed for only one day before approaching Rodriguez with his idea for Krea. The two decided to drop out and build their startup together, turning down the royal scholarships to pursue their vision of making AI image generation accessible to non-technical creators.
Rodriguez applied to HF0, an AI-focused residency program for technical founders in San Francisco. Both were accepted, and Krea was formally founded in March 2022 through the HF0 program. The initial concept centered on helping creative professionals use AI without needing programming skills.
A turning point came when Simian Luo released the Latent Consistency Model (LCM), a technique that dramatically reduced the number of inference steps required for diffusion models. Perez built a live demonstration at a dinner party that transformed real-time camera input into stylized visuals using the LCM approach. This prototype became the foundation for Krea's signature real-time AI canvas, the feature that would set the platform apart from competitors like Midjourney, DALL-E, and Stable Diffusion.
The real-time canvas allowed users to draw simple shapes, add colors, or adjust elements on a digital canvas while the AI generated a fully rendered image in response within milliseconds. This was fundamentally different from the standard workflow of other text-to-image tools, which required users to submit a prompt and wait several seconds (or minutes) for a result.
In March 2024, Fast Company named Krea one of its "Most Innovative Companies." The platform continued to ship new features at a rapid pace, releasing over 50 major product updates since its initial launch. By April 2025, the platform had grown to over 20 million users, and by late 2025, that figure exceeded 30 million users across 191 countries.
Krea's annual recurring revenue (ARR) reached approximately $8 million by April 2025, representing a 20x increase over the preceding 14 months. Enterprise customers began adopting the platform, with notable users including Perplexity, Pixar, LEGO, Samsung, and Loop Earplugs.
Krea's core technical achievement is its real-time generation engine, which produces photorealistic images from simple inputs in under 50 milliseconds. According to co-founder Diego Rodriguez, this speed is achieved through three factors: "(1) using many GPUs, (2) optimization techniques of and around the model, and (3) infrastructure built to scale to millions of users."
The real-time engine uses Latent Consistency Models (LCMs), which reduce the number of denoising steps needed in the diffusion process from dozens to just a few. Traditional diffusion models require 20 to 50 steps to produce a high-quality image; LCMs can achieve comparable quality in 1 to 4 steps, enabling near-instantaneous generation.
For standard (non-real-time) image generation, Krea claims the industry's fastest speeds at 3 seconds for a 1024-pixel Flux image at FP16 precision.
Rather than relying on a single model, Krea aggregates leading generative AI models from multiple providers into one unified interface. As of 2025, the platform supports over 20 image models and multiple video models, including:
| Category | Models Available |
|---|---|
| Image Generation | Flux, Krea 1, Imagen 4, Nano Banana Pro, and others |
| Video Generation | Veo 3, Kling 2.5, Wan 2.5, Hailuo, Runway |
| 3D Generation | Proprietary 3D model (text-to-3D and image-to-3D) |
This approach allows users to select the best model for each task through a single API or interface, without needing separate accounts or integrations with different providers.
In collaboration with Black Forest Labs (the creators of Flux), Krea developed FLUX.1 Krea [dev], a 12-billion-parameter rectified flow transformer for text-to-image generation. The model is architecturally compatible with the FLUX.1 [dev] ecosystem and was trained using guidance distillation for greater efficiency.
FLUX.1 Krea [dev] was trained with the goal of generating more realistic and diverse images that avoid oversaturated textures and the common "AI look" associated with many generated images. In human preference assessments, the model outperforms previous open text-to-image models and matches closed solutions like FLUX.1 [pro].
Key specifications of the Krea 1 model:
| Specification | Detail |
|---|---|
| Parameters | 12 billion |
| Architecture | Rectified flow transformer (compatible with FLUX.1 [dev]) |
| Training Method | Guidance distillation |
| Native Resolution | 1.5K (upscalable to 4K) |
| Generation Time | 7 seconds for HD output with instant previews |
| Supported Aspect Ratios | 2:3, 4:3, 16:9, 2.35:1, 1:1, and more |
| Availability | Open weights via Hugging Face; commercial license through BFL |
The model supports a "Raw" mode for unfiltered output, a preset library with scene and style options, and image prompt and style transfer capabilities. Users describe the output as having natural color grading, cinematic lighting, and diverse style support ranging from photorealistic to surreal to minimalist.
The real-time canvas is Krea's flagship feature. Users interact with a digital canvas by drawing shapes, placing colors, typing text prompts, or even using a live webcam feed. The AI responds instantaneously, generating a fully rendered photorealistic image that updates as the user modifies their input. This turns the creative process into a live, interactive dialogue between the artist and the AI, rather than a prompt-and-wait workflow.
Use cases for the real-time canvas include:
Krea's standard image generation mode supports text-to-image and image-to-image workflows across all integrated models. Users can generate images by typing a text prompt, uploading a reference image, or combining both. The platform provides controls for aspect ratio, style, and model selection.
Krea's upscaling and enhancement tools go beyond simple pixel interpolation. The "Enhance" feature works similarly to Magnific AI, intelligently re-drawing and adding new high-fidelity details as it increases resolution. The platform supports:
| Feature | Capability |
|---|---|
| Image Upscaling | Up to 22K resolution (Max plan) |
| Video Upscaling | Up to 8K resolution |
| Frame Interpolation | Up to 120 fps |
| Basic Upscaling (Free plan) | Up to 2K resolution |
| Standard Upscaling (Basic plan) | Up to 4K resolution |
| Advanced Upscaling (Pro plan) | Up to 8K resolution |
This makes the tool useful for restoring old or low-resolution photographs, preparing images for large-format printing, and enhancing video footage for professional production.
Krea integrates multiple video generation models, allowing users to create AI-generated video clips from text prompts or still images. Supported models include Veo 3 (Google), Kling 2.5, Wan 2.5, Hailuo, and Runway. The "Video Realtime" feature extends the real-time canvas concept to video, producing full-frame consistent video generation.
Video upscaling to 8K resolution and frame interpolation to 120 fps are available for post-processing existing video content.
The AI Patterns feature allows users to generate intricate and repeating pattern designs using text prompts. Users specify a pattern type and description, and the AI produces unique, seamless designs suitable for textiles, wallpapers, packaging, and digital backgrounds.
Logo Illusions transforms a company's logo into a visually striking image that integrates the logo shape into a scene described by a text prompt. For example, the Apple logo could be rendered as a cliffside landscape, with the logo's silhouette embedded naturally into the composition. This feature is designed for brand-related creative work and social media content.
Krea generates 3D objects from text prompts or converts 2D images into 3D meshes. The output objects can be exported for use in 3D software, game engines, or rendered within Krea's own environment. Users can resize and position 3D objects at any angle within AI-generated scenes in real time. When combined with the image upscaler, 3D renders can be converted to photorealistic still images.
Krea Train allows users to upload images (a minimum of 50 on the paid plans) and fine-tune the AI to reproduce specific styles, objects, or faces. This creates a custom LoRA (Low-Rank Adaptation) model that generates images consistent with the user's brand, product line, or artistic style. The feature supports:
| Plan | Training Images Allowed |
|---|---|
| Free | Limited |
| Basic | Up to 50 images |
| Pro | Up to 50 images |
| Max | Unlimited |
| Business | Up to 20,000 images |
Available on the Pro plan and above, Nodes is a visual workflow automation system that lets users chain multiple AI operations together. An AI Nodes Agent can assist in building these automated pipelines, similar to node-based editors in tools like ComfyUI or Unreal Engine's Blueprints.
Krea includes a lipsync feature that synchronizes mouth movements in video with audio input, useful for creating talking-head videos, dubbing content into different languages, or animating still portraits.
Krea provides a REST API that gives developers programmatic access to all of its AI models through a single integration. Key characteristics of the API include:
An interactive API playground is available for testing endpoints and iterating on parameters in a sandbox environment.
Krea uses a freemium model based on compute units. All paid plans include a commercial license for generated content.
| Plan | Monthly Price | Compute Units | Key Features |
|---|---|---|---|
| Free | $0 | 100/day | Real-time models, limited image/video/3D access, 2K upscaling, limited LoRA training |
| Basic | $9/month | 5,000/month | Commercial license, full image/3D/lipsync access, LoRA with 50 images, 4K upscaling, selected video models |
| Pro | $35/month | 20,000/month | All video models, Nodes workflow automation, AI Nodes Agent, bulk discounts, 8K upscaling |
| Max | $70/month | 60,000/month | Unlimited LoRA fine-tuning, unlimited concurrency, unlimited relaxed generations, 22K upscaling |
| Plan | Monthly Price | Compute Units | Key Features |
|---|---|---|---|
| Business | From $200/month | 80,000 to 1.5M/month | Up to 50 included seats, team collaboration, custom roles and permissions, fine-grained model access, LoRA with 20,000 images |
| Enterprise | Custom pricing | Custom | Dedicated support, SAML SSO, analytics API, per-member spend limits, Slack integration, audit logs, custom compute packages |
Annual billing saves 20% across all plans. One-time compute packs with 90-day validity are available in increments of 2,000, 5,000, 10,000, 24,000, and 50,000 units.
Krea has raised a total of $83 million across four funding rounds.
| Round | Date | Amount | Lead Investor | Other Notable Investors |
|---|---|---|---|---|
| Pre-Seed | June 2022 | Undisclosed | HF0 | -- |
| Seed | Early 2023 | $3 million | Pebblebed | A* Capital, Abstract Ventures, Gradient Ventures (Google) |
| Series A | June 2023 | $33 million | Andreessen Horowitz | -- |
| Series B | April 2025 | $47 million | Bain Capital Ventures | Andreessen Horowitz, Abstract Ventures |
The Series B round valued Krea at approximately $500 million post-money. Notable early investors include Pebblebed (founded by Keith Adams, a Meta AI co-founder, and Pamela Vagata, who was on OpenAI's founding team) and Gradient Ventures, Google's AI-focused venture fund.
As of April 2025, Krea employs approximately 17 people, making it one of the leanest AI startups relative to its valuation and user base. The team includes talent from leading universities and researchers who have authored state-of-the-art AI model papers.
The company is headquartered in San Francisco, California. Krea has announced plans to pursue SOC 2 Type II and ISO 27001 compliance certifications for its enterprise product offering.
Krea serves a broad range of creative professionals and industries:
| User Type | Common Applications |
|---|---|
| Graphic Designers | Concept art, marketing visuals, brand asset generation, mood boards |
| Photographers | Image upscaling, photo restoration, AI-enhanced editing |
| Video Editors and Marketers | Animated clips, social media ads, video intros, video upscaling |
| Product Designers | Product photography at scale, 3D mockups, packaging design |
| Game Developers | Concept art, texture generation, 3D asset creation |
| Architects | Design visualization, rendering enhancement |
| Illustrators and Artists | Style exploration, character design, live painting |
Designers using Krea report significantly faster project completion compared to single-model tools like DALL-E 3, largely because of the real-time feedback loop and the ability to switch between multiple models without leaving the platform.
Krea operates in a competitive market that includes both single-model platforms and multi-model creative tools.
| Platform | Developer | Approach | Key Differentiator |
|---|---|---|---|
| Midjourney | Midjourney, Inc. | Closed-source, subscription | Artistic quality, rich aesthetic output, large community |
| DALL-E 3 | OpenAI | Closed-source, ChatGPT integration | Precise prompt following, conversational image generation |
| Stable Diffusion | Stability AI | Open-source, locally runnable | Full customization, local processing, free to use |
| Adobe Firefly | Adobe | Integrated into Creative Cloud | Commercially safe training data, Photoshop integration |
| Runway | Runway | Video-first platform | Leading video generation, Gen-3 Alpha model |
| Leonardo AI | Leonardo Interactive | Multi-model platform | Real-time canvas, model fine-tuning, gaming focus |
Krea differentiates itself from these competitors through its real-time generation speed, its multi-model aggregation approach, and its focus on interactive creative workflows rather than one-shot generation. While Midjourney excels in aesthetic quality and Stable Diffusion offers maximum customization, Krea's core value proposition is the speed of iteration: the ability for a designer to see results change in real time as they modify their input.