Recraft V3 - AI Image Models Tool

Overview

Recraft V3 (codename “red_panda”) is a design-first text-to-image model released in October 2024 that focuses on professional graphic workflows: precise typography, scalable vector output (SVG), and brand-consistent visuals. The model is notable for accurate long-form text rendering (full sentences/paragraphs inside images), deterministic placement of text elements, and a curated + generative style library that preserves style consistency from a few reference images. According to Recraft’s documentation and public announcements, V3 introduced an "Artistic" parameter to tune how conventional versus experimental a composition looks, improved inpainting/outpainting, and first-class support for both raster and vector outputs (SVG) useful for logos, icons, and scalable marketing assets (Recraft docs, Recraft API pages). (https://www.recraft.ai/docs/recraft-models/recraft-V3; https://www.recraft.ai/docs/api-reference/pricing) Recraft V3 gained rapid community attention after anonymously ranking first on the crowdsourced Artificial Analysis text-to-image leaderboard (Hugging Face arena) under the alias "red_panda" and achieving a top ELO score in benchmark comparisons. The model is distributed through Recraft’s Studio (Canvas), a public API, and third-party marketplaces (Replicate, fal.ai and partners), and Recraft allows commercial use of images generated via Replicate and its own API. Independent reviews praise the model’s design controls and vector capabilities; some reviewers note tradeoffs between photorealism and design-centric output depending on style choices (TechCrunch, TechRadar). (https://replicate.com/recraft-ai/recraft-v3; https://techcrunch.com/2024/10/28/a-mysterious-new-image-generation-model-has-appeared/)

Key Features

  • Accurate long-form text rendering inside images (sentences and paragraphs)
  • Native vector (SVG) export for logos, icons, and scalable designs
  • Precise text position and size control (place text at exact coordinates)
  • Brand-style creation: upload references to generate consistent branded assets
  • Artistic-level parameter to adjust fidelity versus experimental composition
  • Inpainting and outpainting for targeted edits and canvas extension
  • Available via Recraft Studio, public API, Replicate and partner marketplaces

Example Usage

Example (python):

import os
import replicate

# Set REPLICATE_API_TOKEN in your environment before running
# export REPLICATE_API_TOKEN="your_token_here"

# Simple example: generate one raster image with Recraft V3 via Replicate
model = replicate.models.get("recraft-ai/recraft-v3")
# Use the latest version; some users call model.versions.get('<version-id>')
# The exact input fields vary by provider; common inputs: prompt, style, size

prompt = (
    "A clean product poster: centered smartwatch on white background, headline 'TIME TO MOVE'"
    ", modern sans-serif, headline at top, small legal text at bottom"
)

output = model.predict(
    prompt=prompt,
    style="realistic_image",   # raster output; use vector_illustration for SVG
    size="1024x1024",
    num_images=1,
)

# model.predict yields a list (batch) of result URLs or data; print first URL
print("Generated output:", output)

# NOTE: depending on Replicate client version you may need to iterate a generator
# or call model.versions.get(...).predict(...). Consult Replicate docs for exact usage.

Pricing

Recraft publishes API-unit pricing: API units are sold at $1 = 1,000 units. Recraft V3 raster generation costs $0.04 per image (40 units); vector generation costs $0.08 per image (80 units). The API docs list per-operation unit charges for inpainting, upscaling, vectorization, and other helpers. (Source: Recraft API pricing docs: https://www.recraft.ai/docs/api-reference/pricing)

Benchmarks

Artificial Analysis (Hugging Face) ELO: 1172 (Source: https://www.recraft.ai/docs/recraft-models/recraft-V3)

Arena win rate (Artificial Analysis): 72% (Source: https://www.techradar.com/computing/artificial-intelligence/the-mysterious-ai-image-generator-that-topped-benchmarks-is-now-available)

Replicate runs (model usage): 7.5M runs (public model page) (Source: https://replicate.com/recraft-ai/recraft-v3)

Last Refreshed: 2026-01-09

Key Information

  • Category: Image Models
  • Type: AI Image Models Tool