AuraFlow v0.3
fal.ai's fully open-source 6.8B flow-based text-to-image model. Up to 1536x1536 resolution.
AuraFlow v0.3 is image generation AI model from Fal.ai, priced at €0.000 per 1M input tokens with a unknown context window.
Pricing
API Integration
Use our OpenAI-compatible API to integrate AuraFlow v0.3 into your application.
npm install railwailimport railwail from "railwail";
const rw = railwail("YOUR_API_KEY");
const images = await rw.run("auraflow-v0-3", "A beautiful sunset over Tokyo");
console.log(images[0].url);
// Or use the image() method for full control
const res = await rw.image("auraflow-v0-3", "A cat in space", {
size: "1024x1024",
n: 1,
});
console.log(res.data[0].url);Deep dive — fal.ai (AuraFlow)'s AuraFlow v0.3
AuraFlow is an open-source flow-matching text-to-image model developed primarily by Simo Ryu (the creator of LoRA fine-tuning) in collaboration with fal.ai (a generative-AI inference platform based in New York, founded in 2021 by Burkay Gur, Gorkem Yurtseven and Batuhan Taskaya). After Stability AI announced restrictive licensing changes to Stable Diffusion 3 in mid-2024, Simo Ryu started AuraFlow as a fully open, transparently licensed alternative. fal.ai sponsored the compute. AuraFlow v0.1 launched in July 2024, with v0.2 and v0.3 (August-September 2024) bringing meaningful quality improvements. The project's stated goal is to give the open community a state-of-the-art flow-matching base that is fully Apache 2.0 licensed and can be freely used and fine-tuned for any purpose.
Visit fal.ai (AuraFlow) →AuraFlow v0.3 is the third public checkpoint in the open-source AuraFlow project led by Simo Ryu and sponsored by fal.ai. The architecture is a Diffusion Transformer (DiT) of 6.8B parameters trained with the rectified-flow / flow-matching objective from the SD3 paper (Esser et al. 2024) and the FLUX.1 family. AuraFlow uses a single T5-XXL text encoder (rather than the dual-encoder design of SDXL or the triple-encoder design of SD 3.5) and operates in the latent space of a learned 16x VAE. AuraFlow was the first fully open-source large-scale flow-matching text-to-image model after the closed-source SD3 release; the v0.3 checkpoint introduced substantial quality improvements over v0.2 thanks to additional pretraining tokens, refined data curation and an updated noise schedule. The model and weights are released under the permissive Apache 2.0 license, including commercial redistribution and derivative works.
- Parameters
- 6.8B parameters
- Context
- 256 tokens
- Fully Apache 2.0 — commercial use, redistribution and derivatives permitted
- 6.8B-parameter rectified-flow DiT — one of the largest open models
- T5-XXL text encoder for strong long-prompt adherence
- 1024x1024 native resolution
- Self-hostable on consumer GPUs with quantisation
- Strong fine-tuning base for community LoRAs and brand models
- Compatible with Diffusers and ComfyUI
- Best for: open community fine-tuning, research, fully unrestricted commercial use, on-prem services.
Trained on a curated subset of publicly available image-text data. Simo Ryu has documented dataset choices and ablations publicly, prioritising transparent provenance over closed data.
License: Apache 2.0 — fully open, commercial use, redistribution and derivative works all permitted.
Known limitations
- Quality below FLUX 1.1 [pro] and SD 3.5 Large
- Smaller training compute than commercial peers
- Hands, text and complex anatomy imperfect
- Open weights have no built-in safety filter
Frequently asked questions
Related Models
View all Image GenerationFlux 1.1 Pro Ultra
FLUX 1.1 Pro in ultra mode. Up to 4 megapixel images with raw mode for photorealism.
Flux Dev
Black Forest Labs' development model. Fast, high-quality image generation with LoRA support.
Google Imagen 4
Google's Imagen 4. Text-to-image with strong photorealism and improved typography support.
Google Imagen 4 Ultra
Premium Imagen 4 tier. Highest fidelity, prompt adherence and typography quality from Google.
Start using AuraFlow v0.3 today
Get started with free credits. No credit card required. Access AuraFlow v0.3 and 100+ other models through a single API.