Recraft launched its AI graphic design platform with native vector (SVG) generation, allowing designers to synthesize scalable icons, logos, and illustrations with editable bezier paths rather than flat raster pixels.
Recraft
Also known as: Recraft V3 · red_panda
Plain English
A graphic-design-focused family of image generation foundation models—famous for topping blind image benchmarks under the codename red_panda and generating true editable vector graphics (SVG) alongside photorealistic images.
Technical Overview
Recraft's proprietary foundation models support both raster and native vector (SVG) geometry generation within an infinite-canvas design environment. Recraft V3 introduced precise bounding-box text placement, long-paragraph typography rendering, and multi-image brand-style conditioning without requiring full LoRA fine-tuning.
Timeline (4)
2023
Release2024
ReleaseRecraft rolled out its proprietary 20B-parameter Recraft V2 foundation model, improving anatomical proportions and multi-icon brand consistency across vector and raster outputs.
- Release
After debuting anonymously as
red_pandato take #1 on the Artificial Analysis Text-to-Image Arena ahead of FLUX1.1 Pro and Midjourney V6, Recraft unveiled Recraft V3 with exact spatial text positioning, long-phrase typography, and native vector generation. 2025
ReleaseRecraft introduced Advanced Style Creation and Controls across its Infinite Canvas and API, enabling teams to blend multiple reference images into reusable brand style tokens without training custom LoRA weights.