Design taste as a first principle
Recraft v4 was built from scratch with professional designers — not just trained on datasets, but built alongside the people who use these outputs professionally. The result is a model that treats composition, color relationships, material realism, and typography as first-class concerns rather than afterthoughts. It outperforms Midjourney V8 and DALL-E 3 by 188 ELO points in head-to-head arena evaluations. The standout capability: native SVG vector output — the only generative model that produces editable, layered vector files directly from a text prompt.Capabilities
#1 benchmark ranking
#1 on the HuggingFace Text-to-Image Arena — ELO 1172 with 72% win rate against Midjourney V8, DALL-E 3, Stable Diffusion, and FLUX.
Native SVG vector generation
The only model in the lineup that generates editable, production-quality SVG files directly — structured layers, scalable to any size without quality loss.
Production text rendering
Reliable, sharp text-in-image generation for infographics, menus, signage, packaging, and UI mockups — with accurate multi-line layout support.
Design-taste composition
Trained with professional designers to produce images with strong compositional logic, cohesive color relationships, and authentic material depth.
Complex prompt adherence
Handles multi-element instructions with minimal element dropout — subject relationships, spatial positioning, and stylistic directives are followed accurately.
Style fidelity
Follows precise design styles — flat illustration, line art, photorealism, brand-aligned aesthetics — with strong consistency across variations.
Specifications
| Feature | Details |
|---|---|
| Resolution | 1024×1024 (1MP) |
| Generation time | ~10 seconds |
| Vector output | Yes (native SVG) |
| Aspect ratios | 1:1, 16:9, 9:16, 4:3, 3:4 |
| HuggingFace Arena | #1 (ELO 1172, 72% win rate) |
| Released | February 2026 |
How to use
Write your prompt
Describe your design intent specifically — style, layout, color palette, subject, and mood. Recraft v4 rewards design-language prompts.
Choose output format
Select raster (PNG) for standard output, or vector (SVG) for scalable, editable files.
Prompting tips
- Use design vocabulary — “flat illustration with muted earthy tones”, “editorial minimalism”, “bold typographic composition” produce stronger results than generic descriptions.
- Text placement instructions work — “Large bold sans-serif text ‘OPEN NOW’ centered, white on navy background, menu design” produces clean, accurate results.
- For SVG vector output — Simpler geometric compositions with clear subject separation produce the cleanest editable vector files. Complex photorealistic scenes work better as rasters.
Example prompts
A minimalist coffee shop menu board with “Morning Specials” as the header, items listed below in clean sans-serif, dark charcoal background, chalk-style illustration accents.
A tech startup logo concept: abstract interlocking geometric shapes, electric blue and white, clean vector style.
A flat-style world map illustration with major cities marked, soft muted palette, clean editorial design.
Compare models
| Model | Resolution | Vector output | Best for |
|---|---|---|---|
| Recraft v4 | 1024×1024 | Yes (SVG) | Design, branding, typography |
| Recraft v4 Pro | 2048×2048 | Yes (SVG) | Print-ready design work |
| Ideogram v3 | Up to 1536px | No | Stylized typography, 58 presets |
Recraft v4 holds the #1 position on the HuggingFace Text-to-Image Arena — outranking Midjourney, DALL-E 3, and FLUX in design-oriented head-to-head evaluations. For print-resolution output, see Recraft v4 Pro.