USE THE NEW EMBEDDING!!!
Completed.
Although the model is far from perfect, this is where this particular experiment ends.
I'll start over with a different training method.
The final version can handle decently 1024x1024 resolution for most of its concepts.
Right now I don't feel like posting the "keywords" the model was trained on, maybe I'll suffice later.
Description
Overall improvement, overfitting in the characters it had learned about, I'll fix it later and provide the list of the characters.
FAQ
Comments (14)
SDXL version?
I don't have the hardware for SDXL, unfortunately.
But if anyone is interested I can provide a part of the dataset, maybe some people can train a SDXL model with my data.
I'm still very much a noob with all of this, I might get back to you once I've figured out how to do it :)
@sandro609 By the way, even if it's not possible for my hardware to train a SDXL model, it is possible to have a SD2.1 in 1024x1024 resolution, and it is in my plans to do so. Are you interested? I might give it more priority if you stick around.
This is actually the first SD2.X model that has caught my interest. I'll play around with it a bit and leave a review with some images once I have some gens I am pleased with. Thanks for your hard work creating this!
It's quite unstable right now, but it has potential. It's recommended to use the embeddings, but they are just shortcuts for complex negative prompts that can be achieved with some patience. Thank you for showing up, I might also fine tune things on request on this model, let me know.
I'll take a pause from this model, I worked all august on this. I hope that the people that can't run SDXL will find something good in ForkRowk, right now it has:
1) better sharpness than most SDXL fine tunings.
2) better prompt understanding than most SD1.5.
3) completely uncensored text encoder, potentially better than most NSFW SD2X on CivitAI
I'll try to train a lora on this model, i have some computing units left on colab and i want to see if it can handle them. If so, we might have the best 2.1 model around.
Looks like this model did not get much attention, I'll soon upload another version but I might as well stop there, seems like it's not worth the trouble.
Details
Files
allworkforkrowk_v02GammaFasterlr512.yaml
Mirrors
providence_2112Anchor.yaml
coloringBook_coloringBook.yaml
Speedtail-000002.yaml
Amg One.yaml
CitroenC6-21-v1.yaml
Speedtail-000004.yaml
Countach lp800.yaml
F1 LM-000004.yaml
ffusionAISD21_768BaseAlpha.yaml
idle_Potion_Generator.yaml
allworkforkrowk_v02GammaFasterlr512.yaml
Orc_reg_128.yaml
Tilt Shift LoHa.yaml
c5x7-21-v01.yaml
AnyTayJoy-woman_Vparam_Reged.yaml
evo_car_D-adaptation-LoRA_SD21-768_v044.yaml
evo_car-dm_plaza_D-adaptation-LoRA_SD21-768_v045.yaml
ffusionAISD21_baseV100.yaml
providence_2111Anchor.yaml
evo-car-lora-sd21-768_v02.yaml
shurima-1-000010.yaml
providence_2110Prelude.yaml
gdsingld.yaml
Aston Martin DBX Interior-000002.yaml
LBWK360MODENA-000001.yaml
m14af1car70s.yaml
drccitycar_v0.3.yaml
Rx7FD-000002.yaml
Speedtail-000001.yaml
Aston Martin DBX-000002.yaml
anime21_v15.yaml
Aston Martin DBX-000001.yaml
Rx7FD-000001.yaml
pixhell_v20.yaml
mcaf1car70s.yaml
DaveScare-v2.yaml



















