Recraft, a San Francisco company developing artificial intelligence for image creation with a specific eye toward brand consistency, has obtained $30 million in Series B financing, the company announced on May 6. The investment round was spearheaded by Accel and saw participation from existing backers Khosla Ventures and Madrona, as well as new investors Nat Friedman, Elad Gil, and RTP Global.
This infusion of capital, which coincides with Recraft’s introduction of “Advanced Style Creation and Controls” for its platform, builds upon a $12 million Series A in 2024. Recraft reports its annual recurring revenue has crossed the $5 million mark, with its user base expanding to 4 million.
The company’s technology, powered by its proprietary “Recraft V3” model, made waves in late 2024. According to a Recraft blog post from October 30, 2024, this model secured the top position on Hugging Face’s Text-to-Image Model Leaderboard by Artificial Analysis with an ELO rating of 1172, outperforming established names like OpenAI’s DALL-E and Midjourney.
Recraft emphasizes that its models are developed in-house and are engineered to provide precise control over the placement and appearance of visual elements, including accurate logo integration and adherence to brand-specific style guides – capabilities often sought after by marketing and design teams.
Recraft V3 is described by the company as a “first-of-its-kind model offering image generation of any size or text length, and precise control over the placement and appearance of visual elements.” The company also claimed in its October 2024 announcement that “Recraft V3 is the only model in the world that can generate images with long texts, as opposed to just one or a couple of words.”
Founder Background and Investor Confidence
Anna Veronika Dorogush, Recraft’s solo female founder and CEO, brings a background in machine learning from Yandex, Google, and Microsoft, and is credited as the creator of the CatBoost library. Before her tech career, Dorogush worked as a professional model while earning a math and computer science degree at a top Russian university.
She told TechCrunch this earlier experience shaped her approach to building the company: “The biggest lesson from that time was that grinding isn’t everything,” Dorogush told TechCrunch. “Now when building a company, I know that to succeed, we have to be excellent at what’s mission-critical. In our case, building models is very important. So we have put all the effort into excelling at this.”
Investors echoed confidence in Recraft’s focused approach. Zhenya Loginov, Partner at Accel, commented: “Anna and the Recraft team are closing the gap between creativity and consistency for design professionals that need to move fast, design at scale, and can’t afford to compromise on control.”
Accel’s own announcement of the investment highlighted the need for professional AI tools to offer not just creative power but also “fine-grained control, be robust and consistent, and support replicable workflows at scale.” The firm also pointed to Recraft’s reported “700% increase in user adoption in the last year alone,” and noted its tools are utilized by teams at organizations including Amazon, NVIDIA, Salesforce, and Uber. Recraft also offers an API, which the company states is used by enterprises needing to generate high volumes of images for advertising, marketing, or other applications.
The Crowded Arena of AI Image Generation
Recraft’s focus on brand-centric design and features like vector image generation (exportable in SVG, PNG, JPG, or Lottie formats) and detailed style controls positions it in a highly competitive field. This includes not only direct AI image generation services like Midjourney, Adobe Firefly, Stable Diffusion, and Black Forest Labs, but also broader design platforms such as Canva, which has itself integrated AI image generation and is an early adopter of OpenAI’s new gpt-image-1 API.
The AI image generation space is characterized by rapid iteration. OpenAI made its GPT-4o based image model, “gpt-image-1,” available to developers via an API on April 24. This allows for programmatic image creation and editing, with OpenAI assuring that API-submitted data is not used for model training.
The service is priced per token, translating to approximately 2 to 19 cents per image. Earlier, in late March 2025, OpenAI had integrated these image generation tools directly into ChatGPT. According to OpenAI, “4o image generation is a new, significantly more capable image generation approach than our earlier DALL·E 3 series of models. It can create photorealistic output. It can take images as inputs and transform them.”
Midjourney, a popular platform known for its artistic output, released an alpha version of its V7 model on April 4, promising improved realism and a “Draft Mode” for faster, lower-cost previews, though with increased computational needs.
Meanwhile, newer services are also vying for market share. Reve Image 1.0, launched March 26, competes on affordability, offering 500 images for $5 and touting strong text rendering capabilities.
ByteDance introduced its Seedream 3.0 model and SeedEdit tool on April 19, targeting global users with a focus on photorealism and accurate text generation, claiming a “94% text availability rate for both Chinese and English characters.”
Industry Considerations and Technical Nuances
Across the sector, addressing the origin and ethical use of AI-generated content remains a key consideration. OpenAI, for example, embeds C2PA (Coalition for Content Provenance and Authenticity) metadata in images from its gpt-image-1 API to denote them as AI-generated, a standard it has supported since early 2024 with DALL·E 3 and further committed to by joining the C2PA Steering Committee in May 2024.
The company also states it avoids mimicking living artists’ styles. These measures align with broader industry discussions and legislative efforts, such as California’s AB 3211 bill, which proposes mandatory AI content labeling.
While Recraft V3 is promoted for its ability to generate images with long text and offer precise control, the general challenge of text rendering consistency remains an area of active development for many AI models.
OpenAI, for instance, notes potential inconsistencies in precise text placement with its gpt-image-1 API, and processing complex prompts can take up to two minutes.
The balance between image quality, generation speed, cost, and the level of user control over output continues to be a defining factor in this evolving market, with the computational resources required by these powerful models also being a background consideration for their widespread adoption and environmental footprint.
