Text-to-Image Magic: FLUX Models Now in Galaxy!
Turn your text into stunning visuals with FLUX models, now integrated into Galaxy!
Using Black Forest Labs' FLUX Models Inside Galaxy: A Guide to Text-to-Image Integration
In previous news, we have written about the speech-to-text tool Whisper and the ChatGPT tool in Galaxy. Today, we would like to talk about a new tool that allows you to create images from text. The Black Forest Labs (BFL) FLUX series has made significant strides in the world of generative AI, providing robust text-to-image capabilities tailored for various use cases. At Freiburg University, we've integrated FLUX.1 [schnell] and FLUX.1 [dev] into the Galaxy platform, enabling researchers and developers to generate high-quality images directly within Galaxy's user-friendly environment.
In this blog post, we'll explore the FLUX.1 models, their unique attributes, and how to use them effectively inside Galaxy.
Overview of FLUX Models
FLUX.1 [schnell]
- Purpose: Designed for speed.
- License: Openly available under the Apache 2.0 license.
- Use Case: Ideal for personal use and rapid prototyping.
FLUX.1 [dev]
- Purpose: Guidance-distilled model for non-commercial applications, offering near-professional quality.
- License: Open-weight model intended for non-commercial contexts.
- Use Case: Suited for research and non-commercial exploration.
Both models provide easy access and versatility, making them excellent choices for text-to-image projects. While FLUX.1 [schnell] focuses on performance, FLUX.1 [dev] bridges the gap between quality and efficiency.
Why Use FLUX Models Inside Galaxy?
Galaxy's open platform supports workflows across life sciences, data analysis, and beyond. By integrating FLUX models, users can:
- Generate visualizations: Convert textual descriptions into detailed images.
- Streamline workflows: Incorporate generative AI into multi-step Galaxy pipelines.
- Leverage open weights: FLUX's availability under open licenses aligns with Galaxy's ethos of accessibility and reproducibility.
To ensure high performance and speed, we are using a CUDA-based container to run the FLUX models on GPUs. This setup allows for fast inference times when working with the models from Hugging Face. By leveraging Hugging Face’s pre-trained models, we can efficiently load the weights and run inference directly within the Galaxy environment. The CUDA-based container ensures that the heavy computations involved in text-to-image generation are handled smoothly, providing users with a seamless experience.
Setting Up FLUX in Galaxy
The FLUX models are already available on UseGalaxy.eu.
Simply log in to your Galaxy instance and search for the FLUX tool in the tool panel.
Select the desired model (FLUX.1 [schnell] or FLUX.1 [dev]) based on your requirements.
Please be adviced that dev model can not be used for commercial use.
Choose between File or Text based input and select your txt file or enter your text prompt in the designated input field. Click "Run Tool" and let Galaxy handle the rest! Once the process is complete, you’ll find the generated image in your Galaxy history, ready to view, share, or use in your workflows.
For example, I used the text prompt "A galaxy full of stars" to generate an image. Here's the result history:
Text Prompt: A galaxy full of stars
Generated Image:
You can try out your own prompts and see what unique images FLUX can generate!