This is more of a test to see how Flux interprets some stuff without giving it to much to work with.Dataset is pretty much only made out of different textures and effects I've created in Photoshop, so no humans, animals, landscapes or anything else that would be recognizable.Since the last commit, Kohya is unfortunately broken for me ( at least FP8 training ). There are a few features that where added which i was waiting for and some i haven't tested yet. Instead of going back to the previous commit i will wait till it's fixed. There is at least one thing i still wanna test and after that a LoRA which i prepared for quite some time and putt a lot of effort into. Not sure if it will work out or not, or if it works if i'm going to release it, but have to wait anyway.<ul><li>Should add Grunge, Fog, Haze, Smoke, Grain, Dust.</li><li>Main goal wasn't really to add those things, more like following a hunch.</li><li>This time, order of the images is with LoRA / without LoRA</li><li>Like always, more like something to play with then anything else. Just judging from the images i assume it works best with drawings. Flux seems to have tendencies for realistic photos ( extremely artificial photos ) or, if drawing, super cartoony. I know that can be negated for the most part by just putting more effort into the prompt, but laziness and exhaustion prevails for now.</li></ul>After all that wanna focus more on images again. Curiosity is the main factor at the moment with everything Flux, which is also extremely time consuming.

Flux.1 D - Image Fubarizer ( Test )

- Flux1 - image_fubarizer_V1.0.safetensors

v1.0

This is more of a test to see how Flux interprets some stuff without giving it to much to work with.

Dataset is pretty much only made out of different textures and effects I've created in Photoshop, so no humans, animals, landscapes or anything else that would be recognizable.

Since the last commit, Kohya is unfortunately broken for me ( at least FP8 training ). There are a few features that where added which i was waiting for and some i haven't tested yet. Instead of going back to the previous commit i will wait till it's fixed. There is at least one thing i still wanna test and after that a LoRA which i prepared for quite some time and putt a lot of effort into. Not sure if it will work out or not, or if it works if i'm going to release it, but have to wait anyway.

    Should add Grunge, Fog, Haze, Smoke, Grain, Dust.

    Main goal wasn't really to add those things, more like following a hunch.

    This time, order of the images is with LoRA / without LoRA

    Like always, more like something to play with then anything else. Just judging from the images i assume it works best with drawings. Flux seems to have tendencies for realistic photos ( extremely artificial photos ) or, if drawing, super cartoony. I know that can be negated for the most part by just putting more effort into the prompt, but laziness and exhaustion prevails for now.

After all that wanna focus more on images again. Curiosity is the main factor at the moment with everything Flux, which is also extremely time consuming.

Files with SHA256: fbea287be485ab1f55032c55ab77b4581a9f86354ae1c8c57ae5d72ab6b62168