Gorgeous 911 GT3 completed with Porsche's new Manthey Performance Kit.
Train your own LoRA with Ostris AI Toolkit on Runpod <--- Referal Link ;)
Description
FAQ
Comments (6)
There's very little Flux Dora made, how would you say it compares with normal Lora?
The challenging part about DoRA is that you need a very high-quality and, above all, diverse dataset. This is because DoRA tends to burn in backgrounds excessively. However, if you have an excellent dataset, I find DoRA to be more advantageous as it is somewhat more detailed and precise.
DoRA is also ideal for quick proof-of-concept (PoC) tests, as you can observe a significant model conversion even after relatively few training steps. In contrast, LoRA requires more steps to achieve similar results.
@denrakeiw Interesting. Thanks for taking the time. Now I'm thinking if dataset masking might help with the background burning in that case.
@razzz Yes, but wouldn’t the mask also burn in? I’ve had quite good experiences when training with 3D data by simply using a different colored background for each image.
@denrakeiw Would have to try. On black it's more less likely to happen. But switching color is a good idea.
Wow !!!
I am back xD
One of my favorite ones !
Amazing !
Could i ask again, why as example the samples if other people are looking that "bad" ?
Are you inferencing on another finetuned flux model?
Or is this vanilla flux?
And how the fuck is it that precised.
Thats amazing !
Youre a god of car training! Love it!
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.