Update V2:
-Did training without faces and backgrounds, flexibility should be better, and faces should be much better
-For now went back to the big file
-For best results use tiriggers "black and grey ds9st uniform" and "teal collar","yellow collar" or "red collar".
V1:
First usable version of XL LORA i made,after many failures,but it still must be improved.
The trigger word is ds9st uniform. When you need a specific color use "black and grey ds9st uniform,teal collar","black and grey ds9st uniform,red collar" or "black and grey ds9st uniform,yellow collar".
The weights i used were 0.8 when using realistic style, but if you want to change you probably have to go down to 0.5.
UPDATE V1.5
-Reduced The file size
-Better flexibility(but still work in progress)
Known issues:
-As stated above the flexibility is still not good enough when adapting styles..
Description
headless version
FAQ
Comments (5)
this fashion is very nice, although when i use the lora (v2) my images turn into a very bad jpg compression look.
will come back for the next version.
Did you try using config like in example images? Or maybe just tell me some basics like sampler and number of steps that you are using so I can test the next version against it.
@impossiblebearcl4060 i tried a lot of configurations/models, also from the examples, but could not get rid of the vhs look. when the lora was deactivated, images were crisp. you can look at my profile or twitter to see what i mean. the vhs look has a charme of course but it would be great if you can minimize it somehow. thank you for the work!
@liia Ah, the film grain. Well, this uniform have the worst data set quality of all my star trek loras, and it shows. But I think I can mitigate it somehow with good captioning, there will be a new version for sure.
i have the same problem :(
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.