Using FLUX 1

Trying out the new FLUX model (FLUX 1 dev).

1. Generated using a prompt found on the net



2. A young woman praying (I created the prompt for this)




I used SwarmUI (ConfyUI in the backend) with the FLUX 1 dev model. Something different, but related to what I studied a long time ago (AI and Computer Science). I have been playing with stable diffusion to generate images, but like many people, the latest SD3 models leave something to be desired. In the last week or so, the original developers of stable diffusion released the FLUX 1 set of models. It was time to give it a try, and these images are the results of trying the FLUX 1 dev out for a few hours. The technology is very interesting. The only question is what images and other data do people use to create these models. No one really knows.

I have also been playing with the llama 3.1 70B model released by Meta. This is quite interesting. I am hoping to try to figure out how well this model works with extracting entities and other information from text, but for now, I am happily testing out it's chat GPT like capabilities using openllama.

I think there is a long way yet before these models will actually understand text in a contextual manner in a broad domain. Still, a lot has happened in the last few years and it's great that Meta released llama 3.1 for people to use. I just hope these technologies are put to good use. We shall see.

In the meantime, it's really interesting to see how several lines of text can be used to generate interesting images. With FLEX, the images at least deviate from the ones that SDXL and models based on it are able to generate (the people generation tend to result in similar looking people).   

Thanks for looking. And for putting up with the computer stuff.

Edited by Almazar80 - 09 August 2024 at 04:29

Flux is used when soldering! Sorry, couldn't resist!

Ah, ha, thank you Winston! And your photos are very nice! Beautiful portrait "A young woman praying"! TFS!

Kelly

The built in video in my Mac Mini M1 might not qualify for Flux!

While these pictures are not really bad they are still easily recognised as AI generated.
My two biggest problems with AI generated images are:

First, they seem way overprocessed and too clean to be real photos.

Secondly there's no guarantee that all owners of the "real" photos to train the AI model and the people pictured in those photos have given their permission to use these photos for that purpose.

I use to think Aldous Huxley had a more persuasive presentation of the future than George Orwell but the last 20 years have convinced me that Orwell had it right all along. AI imaging programs like this are a Ministry of Truth wet dream and a boon for child pornographers everywhere.