Recraft V3, codenamed red_panda, is a 20-billion parameter text-to-image model released in October 2024. Built from the ground up for professional design workflows, it's the first AI image generator that actually thinks in design language rather than just interpreting text prompts.
Since its launch, Recraft V3 held the #1 position on the Artificial Analysis Text-to-Image Leaderboard on Hugging Face for over five consecutive months, outperforming Midjourney v6.1 (1093 ELO), FLUX 1.1 Pro (1143 ELO), and OpenAI's DALL-E 3 HD (984 ELO) with an ELO rating of 1172 and a 72% overall win rate.
Every other AI image model struggles to render even a few words correctly. Recraft V3 is the only model in the world that can generate images containing full sentences, multi-line copy, headlines, and paragraph text — clean, readable, and properly typeset. Build ad banners, posters, product packaging, and social graphics without touching Photoshop.
Stop wrestling with vague prompts and hoping the model figures out your layout. Recraft V3 lets you specify exact positions and sizes for text elements within the image. Place a headline at the top, a product name across the center, and a call-to-action at the bottom and get exactly that.
Midjourney, DALL-E, and most other models only produce raster images. Recraft V3 generates true scalable vector graphics (SVG) natively, from clean icon sets and minimal pictograms all the way to intricate illustrations. Perfect for logos, app icons, brand assets, and anything that needs to scale from a favicon to a billboard.
Upload a few reference images and Recraft V3 locks in your brand's visual style — color palette, composition feel, artistic tone — and maintains it across every generation. No model retraining, no LoRA workflows, no expensive fine-tuning jobs
Fingers, hands, limbs — AI anatomy has always been a punchline. Recraft V3 was specifically trained on anatomical accuracy metrics, producing correctly proportioned bodies, natural hand poses, and spatially coherent scenes. Professional lifestyle photography and product mockups that actually look real.
Recraft V3 handles complex, multi-object scenes with correct count, accurate colors, and proper object positioning, precisely as described. Object relationships, background depth, lighting conditions, the model tracks the details other models ignore. Less iteration, more first-pass accuracy.
Pass exact hex color codes directly in your API request and Recraft V3 generates images using those precise brand colors. No more color drift between generations. Every asset stays on-brand without manual color correction.
Recraft V3 includes inpainting, outpainting, AI Eraser, background removal, creative and crisp upscalers, AI mockup generation, and image vectorization.
| Feature | Recraft V3 | Midjourney v6 | DALL·E 3 HD | FLUX 1.1 Pro |
|---|---|---|---|---|
| HuggingFace Benchmark Rank | #1 (1172 ELO) | #3 (1093 ELO) | #5 (984 ELO) | #2 (1143 ELO) |
| Long text in images | ✓ Full sentences | ✗ 1–2 words max | ✗ Often garbled | ✗ Very limited |
| Text position control | ✓ Exact | ✗ | ✗ | ✗ |
| Native SVG vector output | ✓ True SVG | ✗ Raster only | ✗ Raster only | ✗ Raster only |
| Brand style (no fine-tuning) | ✓ From examples | ✗ | ✗ | ✗ |
| Hex-accurate brand colors | ✓ | ✗ | ✗ | ✗ |
| Commercial use | ✓ | Plan-dependent | ✓ | ✓ |
The Artistic Level parameter lets you dial creativity from 0 (straightforward portrait, literal interpretation) to 10 (unconventional angles, experimental compositions). This gives developers fine-grained control over aesthetic output without switching models.
The combination of text rendering, vector output, and brand control opens up use cases that were impossible with previous generation models.
You can compare the images from this model with those from Stable Diffusion 3.5 in this blog post Curious About Stable Diffusion 3.5 Large?
Measured on the AI/ML API gateway across live traffic. Figures are medians unless stated. Last updated 2026-10-12.
import requests r = requests.post( "https://api.aimlapi.com/v1/images/generations", headers={"Authorization": "Bearer " + AIMLAPI_KEY}, json={ "model": "recraft-v3", "prompt": "A cat holding a sign that says hello" }, ) print(r.json())
const r = await fetch("https://api.aimlapi.com/v1/images/generations", { method: "POST", headers: { Authorization: `Bearer ${process.env.AIMLAPI_KEY}`, "Content-Type": "application/json", }, body: JSON.stringify({ "model": "recraft-v3", "prompt": "A cat holding a sign that says hello" }), }); console.log(await r.json());
curl -X POST https://api.aimlapi.com/v1/images/generations \ -H "Authorization: Bearer $AIMLAPI_KEY" \ -H "Content-Type: application/json" \ -d '{"model":"recraft-v3","prompt":"A cat holding a sign that says hello"}'
OpenAI-compatible — swap the base URL and it works with your existing SDK.
| Type | Price |
|---|---|
| Output | |
| Benchmark | Score | What it measures | Source | Retrieved |
|---|---|---|---|---|
| Text-to-Image Arena | 868 | Human preference Elo from blind pairwise comparisons of generated images, measured independently by Artificial Analysis | Source | September 28, 2026 |
| Model | Input | Output | Context | Best for |
|---|---|---|---|---|
Recraft V3 This page | Image generation at scale | |||
| Image generation at scale | ||||
| Image generation at scale | ||||
| Image generation at scale | ||||
| Image generation at scale |
Recraft V3 takes image, text as input and returns image.
Recraft V3 is priced at $0.052 / gen (variable).
Recraft V3 is billed per generation — a fixed charge per output rather than by prompt length.
Yes, Recraft V3 accepts image input as well as a text prompt.
Recraft V3 was built by RecraftAI.
Send a request to https://api.aimlapi.com/v1/images/generations with recraft-v3 as the model id.
Yes. Recraft V3 is served through AI/ML API, so the same key and endpoint format used for other models applies.
No. Recraft V3 is an image generation model that converts text into images, not a general-purpose multimodal LLM.
Recraft V3 generates images from text prompts and can produce vector-style graphics.
