August 2024: Black Forest Labs Releases Flux.1 Models
New Models and Features
On August 1, 2024, Black Forest Labs released three new models, Flux.1 – Pro, Dev, and Schnell. The Pro version is not open source and is available through their API, while DEV and Schnell are both open source and available to download via Huggingface page.
Model Comparison
The DEV model is a higher quality model than Schnell, but Schnell is much faster, with a generation time of 4 steps. Both models are large, weighing in at 23.8GB each, and require a high level of VRAM to run. It is recommended that you have 32GB RAM. However, it is possible to run them on lower VRAM GPUs, as demonstrated by the author, who uses an RTX4080 with 16GB of RAM.
Comparison to Midjourney
The buzz around these models is that they are at par with Midjourney, and in the author’s testing, they agree that they are much better. The Flux.1 models excel in several areas, including:
- Resolution: The model can handle any image size, from extremely wide to extremely tall, with no set resolution required.
- Prompts: The model is much better at handling prompts and adhering to the nuances of the prompt.
- Quality: The quality of the generated images is higher, with better-formed hands, composition, and facial features.
- Text: The model renders text better than any other model, even SD3.
What Sets Flux.1 Apart
One of the key differences between Flux.1 and Midjourney is that Flux.1 doesn’t apply its own recipe or sauce to make the image better, staying closer to the original prompt. This makes it easier to control the output with a text prompt.
Downloading and Running the Models
To run the Flux.1 models, you need ComfyUI (update to the latest version) and then download the necessary files. Place the Model in the models\unet folder, VAE in models\VAE, and Clip in models\clip folder of ComfyUI directories. Make sure to restart ComfyUI and refresh your browser.
Workflows and Sample Results
The author provides two workflows, Text 2 Image and Image 2 Image, which can be downloaded and used with ComfyUI. The Image 2 Image workflow utilizes Florence 2 LLM and Clip Interrogator to generate an accompanying prompt to help guide Flux.
Sample Results
The article includes several sample results from the Flux.1 models, showcasing their high-quality and coherent output.
Conclusion
The release of Flux.1 models has been a welcome addition to the AI art community, offering high-quality and coherent results. While it’s still early days, the author is excited to see what other developments await us in the coming months.
FAQs
Q: What is the difference between Flux.1 and Midjourney?
A: Flux.1 doesn’t apply its own recipe or sauce to make the image better, staying closer to the original prompt, making it easier to control the output with a text prompt.
Q: Can I run Flux.1 on a lower VRAM GPU?
A: Yes, it is possible to run Flux.1 on a lower VRAM GPU, but it may take longer to generate images.
Q: What is the recommended RAM for running Flux.1?
A: 32GB RAM is recommended, but it is possible to run it on lower RAM configurations.
Q: Can I download the Flux.1 models?
A: Yes, the DEV and Schnell models are available to download via Huggingface page.
