OpenFLUX.1: The New AI Image Model That Rivals FLUX.1[dev]

The world of AI-generated images keeps getting better, and now we’ve got something exciting: OpenFLUX.1! It’s an open-source version of the popular FLUX.1-schnell model, but it’s taken a big step forward by removing “distillation,” making it more flexible. This change means you can fine-tune it and create top-notch images with just a few tweaks.

In this article, we’ll look at how OpenFLUX.1 stacks up against other models, including its earlier versions, and see what makes it so special.

1. What is OpenFLUX.1?

OpenFLUX.1 is a fine-tuned version of FLUX.1-schnell without the distillation. Distillation in the original FLUX.1-schnell made it a great model, but you couldn’t fine-tune it. That limited its flexibility. However, OpenFLUX.1 changes all that and allows you to make more adjustments, perfect for those who want to get creative. Even with these changes, it keeps the ability to generate stunning images in just 1 to 4 steps.

The model comes under an open-source license, giving users more freedom to experiment and improve it. Think of it as a tool for both professionals and hobbyists who want more control over their image generation projects.

https://huggingface.co/ostris/OpenFLUX.1

2. Getting Started with OpenFLUX.1 and ComfyUI

Using OpenFLUX.1 with ComfyUI is super easy. The developers have even made FP8 weights (a type of optimized model file) for it. You can download these and drop them into your ComfyUI folder like this:

Download the OpenFLUX.1 model and place it in the
```
ComfyUI/models/checkpoints
```
folder.
Grab the FP8 version workflow from the ComfyUI Examples link provided on their page.
You’re all set to start creating stunning images!

It’s really just that simple. No complicated steps or confusing setups. It’s designed to be user-friendly, especially if you’re familiar with ComfyUI already.

https://huggingface.co/ostris/OpenFLUX.1/blob/main/openflux1-v0.1.0-fp8.safetensors

https://comfyanonymous.github.io/ComfyUI_examples/flux/#simple-to-use-fp8-checkpoint-version

3. Testing the Results

Now for the fun part—let’s compare some actual generated images between OpenFLUX.1 and the older FLUX.1[dev].

Here’s what we used for the test:

Seed: 232145193935489
CFG (a setting that controls how closely the model follows your instructions):
- OpenFLUX.1: 3.5
- FLUX.1[dev]: 1.0
Steps:
- OpenFLUX.1: 30 steps
- FLUX.1[dev]: 20 steps
Sampler: Euler
Scheduler: Simple

Visit assets.st-note.com

We tried out various prompts and here’s what stood out:

Quality: Both models produce similar compositions, but OpenFLUX.1 shows a clear improvement over the original Schnell model in terms of sharpness and overall image clarity.
Fidelity to Prompts: OpenFLUX.1 follows prompts much better than the older model. For example, we used the prompt “A group of adventurers consisting of a knight, a mage, and a rogue…” and while FLUX.1[dev] nailed the details, OpenFLUX.1 did an impressive job too, though it sometimes added an extra character.
Handling People: OpenFLUX.1 can handle drawing multiple characters, but when the scene gets too complex or zoomed out, things get a bit wonky, especially with hands and character proportions.

Even though dev still holds its ground in some aspects, OpenFLUX.1 is a worthy competitor. It’s especially useful for those who want to fine-tune and have more control over their image outputs.

Summary

In a nutshell, OpenFLUX.1 is a game-changer for AI image generation. It’s flexible, fine-tunable, and open-source, giving you more freedom to tweak and improve your image generation process. Whether you’re a professional looking to push the boundaries or a casual user looking for more control, this model has a lot to offer.

OpenFLUX.1 is likely to be a huge player in the future, especially with its ability to evolve through community-driven improvements. Get ready for more updates and even better results as more people start fine-tuning it for their own projects!

In the end, OpenFLUX.1 look at the future of image generation, and it might just signal the end of models like SDXL. Stay tuned!