Pretty self-explanatory for anyone who knows it. Does not include character styles, only backgrounds.
Description
FAQ
Comments (10)
Curse of Monkey Island?
Indeed :)
I've tried this Lora on one of those paying websites where I could import the Lora if I was a paying member and it worked really good, almost as good as your results. However I want to use it in automatic1111 locally and I can't get it to work properly despite all my efforts. I can see some influence of the Lora on the images I think but its not even close to the results I had on the other site. Do you have any tips on how to get it to work? I'm using the model Juggernaut XL and DPM++ 2M Karras and CFG scale 6. Am I missing something?
Try upping CFG to 9-10, and set the lora influence to 1.0, that should help.
@rb_ thanks for the quick reply and Lora! It was such a good nostalgy trip I even installed CMI ahah. However sadly Im still not getting the results. I tried fiddling with CFG scale but it didnt produce any good results. What is strange is if I use Pixart.ai website I get great results. I tried copying the negative prompt from there to the auto1111 but still it was really lackluster. Could you give me a prompt example and the negative prompt also you used in one of the images you generated?
@sf6world None of the prompts I used for the examples were doing anything special. Most of them were something along the lines of "COMIstyle style of [random_location_here]" with nothing in negative. Maybe try using the base SDXL model instead of Juggernaut, that might be the problem.
@rb_ I see... well I will give up for now. Tried the base SDXL model too. But thanks for the help!
@sf6world there might be something wrong with your 1111 install as well, have you tried using Fooocus?
@rb_ I did not know about Fooocus. Just installed it and couldn't get results either. Used the prompt 'pyramid, desert, entrance to a crypt, COMIstyle'. On the Styles left the default Fooocus v2, Fooocus Enhance and Fooocus Sharp. On the Lora left the default sd_xl_offset_example-lora_1.0.safetensors at 0.1 and COMIstyle at 1. Used the resolution 1280x768. Tried different guidance scales. The results were not even close to yours :(
@sf6world You need to disable v2, enhance and sharp - those add way too much stuff to the prompt to make the lora work, also - try moving COMIstyle so the beginning of the prompt, and then it should work, that's pretty much what I did for those example images.