Krea raises Series B with 20 million users — co-founder Víctor Perez on AI image generation adoption and the OpenAI image model breakthrough

Summary

Víctor Perez, co-founder of Krea, closed a Series B with 20 million users on the platform. Krea's user base splits into three groups. Consumers with no prior creative background pay for subscriptions the way they'd pay for a camera or video game, gaining a creative capability they didn't have before. Professional creatives use it to compress production workflows: architecture studios upload low-resolution renders and upscale them to 4K, while game designers use Krea's real-time tool to rapid-prototype character concepts. Enterprise represents the third group, though Perez does not break out its size.

Consumer awareness is still very early

The AI image generation market is barely penetrated. Perez returned from New York noting that the San Francisco bubble makes it easy to overestimate how widely this technology is understood. Most people don't know AI image generation exists, let alone that it could solve a marketing or product design problem. An Uber driver in New York selling beauty products on Instagram was immediately interested in using Krea for product photography. The workflow remains a multi-step process that requires someone to explain it: train a model on your product, then generate assets.

Prompt engineering as software development

Prompt engineering is not a narrow job title but a new mode of building software. Most future software, including Krea itself, will be created increasingly through natural-language instructions to AI models rather than traditional coding. The visual creative workflow is shifting the same way: users will steer models through instructions rather than manually tuning image parameters.

OpenAI's image model

The new OpenAI image model demonstrates that shift most clearly. Its differentiator is not rendering quality alone but the model's ability to reason about instructions. Prior diffusion models could translate text into an image that roughly matched the description but struggled with multi-step instructions like "here's a photo of my dog, turn it into Studio Ghibli." The new model handles that. Perez suspects an autoregressive architecture is involved, pointing to how the image renders top-to-bottom rather than diffusing globally from blurry to crisp, and notes similarities to Anthropic's image generation approach. He does not have high confidence in that technical read. The OpenAI release validates where Krea is headed: a platform built around instructable, reasoning-capable image models rather than prompt-and-generate diffusion pipelines.