A few days ago Flux Redux was released as part of the FLUX.1 Tools, by Black forest Lab. The complete list of this release are as follows.
- FLUX.1 Fill: State-of-the-art inpainting and outpainting models, enabling editing and expansion of real and generated images given a text description and a binary mask.
- FLUX.1 Depth: Models trained to enable structural guidance based on a depth map extracted from an input image and a text prompt.
- FLUX.1 Canny: Models trained to enable structural guidance based on canny edges extracted from an input image and a text prompt.
- FLUX.1 Redux: An adapter that allows mixing and recreating input images and text prompts.
The first 3 in the list isn’t something new really, but have been available for a long time for a variety of models. Flux Redux might not be a completely new concept either, as there has been many different methods working to achieve the same thing – Enhance, alter and improve already existing images.
Image-to-image (IMG2IMG) is probably the earliest method, and it’s still today a very powerful tool for altering an image you already have. Inpainting and outpainting is 2 other methods that can be used for the same purpose, and then there is Gligen. I have personally never used Gligen, but the way I understand it, it’s kind of a more accurate method to inpaint (I might be wrong though).
Just recently I started using a method to manipulate the sigma, making the model to produce slight differences between the output images, even if using the same seed and settings. For me that’s a perfect way to both get variation, but also find new inspiration. The workflow I’m using for that looks like this.
So I use a single prompt and a single seed and get 4 images that are similar, but still have enough variation between them to make it interesting.
Flux Redux
Flux Redux can be used in different ways, all from a very simple workflow to alter a phot into an anime image, to huge workflows that creates a stream of images with lesser or larger variations.
FLUX.1 Redux is an adapter for all FLUX.1 base models for image variation generation. Given an input image, FLUX.1 Redux can reproduce the image with slight variation, allowing to refine a given imageIt naturally integrates into more complex workflows unlocking image restyling via prompt. Restyling is available through our API by providing an image plus a prompt. The feature is supported in our latest model FLUX1.1 [pro] Ultra, allowing for combining input images and text prompts to create high-quality 4-megapixel outputs with flexible aspect ratios.
I wasn’t very impressed by my first few attempts. But at the time I used the very simplest of workflow, and didn’t really know what all the different settings was for (I still don’t).
These were some of the first images I created, and my first thought was that I can do this myself, with just regular prompting. But then I got a workflow from a guy in an AI related Facebook group that I wanted to try. I wasn’t completely happy with it, so I made some small adjustments. Here is the result (click on the image to make it larger).
This workflow managed to make the image to the left below, into the image to the right. And that’s pretty impressive in my book.
You can download the workflow here if you want to try it out: Acid Flux Redux – Fixed
I wouldn’t recommend that big workflow if you have less than 10GB VRAM and 32GB system RAM though. A computer with less than that might be able to handle it, but it will put a lot of strain on your system and take a very long time to finish.
If you have less VRAM and RAM than stated above, you might want to try the simple workflow, or the semi advanced one instead.
Download them here:
Flux Redux – Semi Advanced Workflow
Have fun!
And sign up for my irregular newsletter, to get mini-updates like these directly in your inbox.