Recraft profile, products and governance

The product most people meet is Recraft, a browser design canvas with a vector generator, an image vectorizer, a mockup generator, an upscaler, a background remover and an eraser, plus iOS and Android apps. Studio also hosts third-party models next to Recraft's own: its model list names Nano Banana, Flux, Ideogram, GPT Image, HiDream, Imagen, Qwen Image, Seedream and Grok Image for images, and Veo, Kling, Sora, Seedance, Pixverse, Hailuo, Wan and Ray for video, so the FLUX family from Black Forest Labs can be picked in Studio alongside Recraft's own models. Developers get a REST API (with an API Portal for usage and tokens, launched on 10 November 2026) and an MCP server, announced on 14 July 2025, that lets Claude, Cursor and other agents call the same tools. Opened on 12 November 2026 without an account, Recraft's public community gallery offered filters for photorealism, illustration, vector art, graphic design and logo; the images themselves were not assessed.

Dorogush said in May 2025, as relayed by Dataconomy, that Recraft develops its models from scratch. Recraft V3, codenamed red_panda, topped the Artificial Analysis benchmark in 2024, according to the same Dataconomy report. Recraft V4 followed on 17 February 2026, and Recraft V4.1 on 14 May 2026 in three modes: the main model, Utility for more predictable production output, and Vector for native SVG. Recraft's changelog entry of 23 September 2026 says V4.1 Flash generates an image in 1.3 seconds end to end, a vendor-measured figure. Recent product moves include Agentic mode (8 December 2025), Exploration Mode (25 March 2026), the Vector Editor (9 July 2026), V4 Styles (26 August 2026), Logo Maker (3 November 2026) and Vectorizer Pro (11 November 2026). Recraft's own 30 May 2026 press release now calls it an independent AI lab, and says its research team is fewer than 20 people.

Recraft announced a $12M Series A on 18 January 2024, co-led by Khosla Ventures and Nat Friedman, with RTP Global, Abstract VC, Basis Set Ventures, Elad Gil and angel investors. The $30M Series B came on 1 May 2025, led by Accel with Madrona Ventures and the existing investors, and a 28 July 2025 extension added NVIDIA and Ethan Eismann, listed as SVP Design at Slack, with no amount given. No later round and no valuation has been announced by the company. TechCrunch, as relayed by Dataconomy in May 2025, reported more than $5M in annual recurring revenue and 4 million users at that point; Recraft's About page now says more than 3 million users in 200 countries and more than 350 million images generated, without a date.

Recraft sells paid Studio plans that include credits and, separately, prepaid API Units that do not expire and are non-refundable, per its pricing page. Its customer logos include HubSpot, Airbus, Asana, Ogilvy and Miro, and the Series B release said teams at Amazon, NVIDIA, Salesforce and Uber use it.

Recraft holds a SOC 2 Type II certification effective 18 December 2025 and announced AIUC-1 certification on 22 December 2025, a standard for testing AI behavior in production; its trust center also lists GDPR and CCPA.

Where Recraft is weaker is raw image ranking. On Artificial Analysis's blind-vote leaderboard, OpenAI's GPT Image line and Google's Nano Banana models from Google DeepMind sit above Recraft's V4.1 models as of 12 November 2026, so a buyer whose only test is photoreal preference scores should compare those first. Dorogush names Midjourney, Adobe Firefly, Stable Diffusion and Black Forest Labs as the image generators it competes with, and the same report adds Canva as a design-tool rival for branded graphics; see Midjourney, Adobe Firefly and Canva. Recraft's case rests on editable SVG output and saved brand styles, which suits designers, print-on-demand sellers and marketers more than people who want a one-off picture. Its startup profile sits with the other venture-backed AI companies on Hokai.

Choose Recraft when the job ends in an editable SVG, a logo or a brand-consistent set of assets. Start with OpenAI or Google DeepMind instead if you want the highest preference-ranked photoreal images, and use the API or a Group Workspace plan if your content is sensitive, since the privacy policy keeps those two out of Recraft's own model training.