aquiles-image 0.1.90

Aquiles-Image.

Aquiles-Image

Easy, fast and cheap Diffusion Models that work for everyone.

🚀 FastAPI • Diffusers • Compatible with the OpenAI client

🔥 What's New in Aquiles-Image

Feature	Description
⚡ 3x Faster	Advanced inference optimizations
🎨 More Models	Support for FLUX, SD3-3.5, Flux2 and more
🔧 Better DevX	Improved CLI and monitoring capabilities
🔌 OpenAI Compatible	Drop-in replacement for OpenAI's image APIs

📋 Prerequisites

• Python 3.8+
• CUDA-compatible GPU with 24GB+ VRAM
• 10GB+ free disk space

Generating an image with stabilityai/stable-diffusion-3.5-medium

https://github.com/user-attachments/assets/00e18988-0472-4171-8716-dc81b53dcafa

Generating an image with black-forest-labs/FLUX.1-Krea-dev

https://github.com/user-attachments/assets/00d4235c-e49c-435e-a71a-72c36040a8d7

Editing an image with black-forest-labs/FLUX.1-Kontext-dev

Edit Script	Result

⚙️ Installation

From Pypi

uv pip install aquiles-image

From source

git clone https://github.com/Aquiles-ai/Aquiles-Image.git
cd Aquiles-Image
uv pip install .

🚀 Launch your Aquiles-Image server

aquiles-image serve --host "0.0.0.0" --port 5500 --model "stabilityai/stable-diffusion-3.5-medium"

🎨 Supported Models

Model	Endpoint
stabilityai/stable-diffusion-3-medium	/images/generations
stabilityai/stable-diffusion-3.5-medium	/images/generations
stabilityai/stable-diffusion-3.5-large	/images/generations
stabilityai/stable-diffusion-3.5-large-turbo	/images/generations
black-forest-labs/FLUX.1-dev	/images/generations
black-forest-labs/FLUX.1-schnell	/images/generations
black-forest-labs/FLUX.1-Krea-dev	/images/generations
diffusers/FLUX.2-dev-bnb-4bit	/images/generations
black-forest-labs/FLUX.2-dev	/images/generations
Tongyi-MAI/Z-Image-Turbo	/images/generations
black-forest-labs/FLUX.1-Kontext-dev	/images/edits

⚠️ VRAM Requirements: Most models require 24GB+ VRAM with an additional ~10GB free for processing.

🧪 Experimental: AutoPipeline Support

Aquiles-Image now supports running additional models through the experimental AutoPipeline feature. This allows you to use any model compatible with AutoPipelineForText2Image from Hugging Face's Diffusers library.

What is AutoPipeline?

AutoPipeline automatically detects and loads compatible diffusion models without requiring manual configuration. While this provides greater flexibility, it comes with some trade-offs.

Supported Models

Any model that can be loaded with AutoPipelineForText2Image without extra configurations, including:

• stable-diffusion-v1-5/stable-diffusion-v1-5
• stabilityai/stable-diffusion-xl-base-1.0
• And many more from HuggingFace Hub

Usage

aquiles-image serve --host "0.0.0.0" --port 5500 \
  --model "stabilityai/stable-diffusion-xl-base-1.0" \
  --set-steps 30 \
  --auto-pipeline

Example with OpenAI Python Client

from openai import OpenAI
import base64

client = OpenAI(base_url="http://127.0.0.1:5500", api_key="__UNKNOWN__")

result = client.images.generate(
    model="stabilityai/stable-diffusion-xl-base-1.0",
    prompt="a beautiful sunset over mountains",
    size="1024x1024"
)

print(f"Generated image: {result.data[0].url}")

# Download the image
image_bytes = base64.b64decode(result.data[0].b64_json)
with open("output.png", "wb") as f:
    f.write(image_bytes)

More Examples

Using Stable Diffusion v1.5:

aquiles-image serve --model "stable-diffusion-v1-5/stable-diffusion-v1-5" --set-steps 40 --auto-pipeline

Using SDXL Base:

aquiles-image serve --model "stabilityai/stable-diffusion-xl-base-1.0" --set-steps 30 --auto-pipeline

With custom port and API key:

aquiles-image serve \
  --model "stabilityai/stable-diffusion-xl-base-1.0" \
  --set-steps 30 \
  --auto-pipeline \
  --port 5500 \
  --api-key "your-secret-key"

⚠️ Important Limitations

Limitation	Description
🐌 Slower Inference	AutoPipeline models may have longer inference times compared to optimized native implementations
🚫 No LoRA Support	LoRA adapters are not currently supported
🚫 No Adapters	Other adapter types (ControlNet, T2I-Adapter, etc.) are not supported
🧪 Experimental	This feature is in active development and may have stability issues
📦 Limited Configs	Only models that work out-of-the-box with default AutoPipelineForText2Image settings

Troubleshooting

If you encounter errors with AutoPipeline:

1. Check model compatibility: Ensure the model supports AutoPipelineForText2Image

   # Test locally first
   from diffusers import AutoPipelineForText2Image
   pipe = AutoPipelineForText2Image.from_pretrained("model-name")
   

2. Verify VRAM: Some models may require more VRAM than expected

   • Check model card on Hugging Face for requirements
   • Monitor GPU memory usage during inference

3. Use native implementations: For supported models (FLUX, SD3, etc.), use the optimized native pipelines instead

   # Instead of --auto-pipeline, use native support 
   aquiles-image serve --model "stabilityai/stable-diffusion-3.5-medium"
   

4. Check logs: Enable verbose logging to see detailed error messages

   # The server logs will show detailed error information
   # Look for errors starting with "X" emoji
   

💡 Tip: If a model is frequently used in your workflow, consider requesting native support by opening an issue on our GitHub repository.

Reporting Issues

When reporting AutoPipeline issues, please include:

• Model name and HuggingFace URL
• Full error message and logs
• GPU model and VRAM amount
• Aquiles-Image version (pip show aquiles-image)

Create an issue at: https://github.com/Aquiles-ai/Aquiles-Image/issues

💻 Start your Aquiles-Image server in dev mode without loading models

Dev mode allows you to start the server without loading any AI models, ideal for rapid development, integration testing, or endpoint validation without requiring GPU or heavy computational resources.

aquiles-image serve --host "0.0.0.0" --port 5500 --no-load-model

What does dev mode do?

• No model loading: Server starts instantly without downloading or loading AI models
• Functional endpoints: All endpoints respond normally with test images
• Realistic responses: Returns valid images that simulate model responses
• Same format: Responses maintain the exact API format (URLs, base64, metadata)

Use cases

• API integration development
• Endpoint testing without GPU
• Workflow validation
• CI/CD environment testing
• Development on machines without GPU resources

Note: Dev mode is for development only. For production, use the normal server with loaded models.

🎉 Generate your first image with Aquiles-Image

from openai import OpenAI
import requests

client = OpenAI(base_url="http://127.0.0.1:5500", api_key="__UNKNOWN__")

result = client.images.generate(
    model="stabilityai/stable-diffusion-3.5-medium",
    prompt="a white siamese cat",
    size="1024x1024"
)

print(f"URL of the generated image: {result.data[0].url}\n")

image_url = result.data[0].url
response = requests.get(image_url)

with open("image.png", "wb") as f:
    f.write(response.content)

print(f"Image downloaded successfully\n")

🎯 Perfect For

Use Case	Description
🚀 AI Startups	Building image generation features
👨‍💻 Developers	Prototyping with Image Generation Models
🏢 Enterprises	Scalable image AI infrastructure
🔬 Researchers	Experimenting with multiple models

Built with ❤️ for the AI community

⭐ Star this project • 📖 Documentation • 💬 Community