yes you need multipler of 64
yes you need multipler of 64
Fthat is limitation
Falso you need to increase max allowed size
FGitHub
I had to modify this code to generate TensorRT that will support highres fix for SDXL self.max_image_shape = 768 if version in ("1.4", "1.5") else 2048 It is by default 1024 So ...
LI tried both but my regular sizes engines are getting errors as well so it's something else. My venv is Python 3.10.6 do I need to update?
PI'm horrified seeing the quality of A1111's python code and its extensions, even more so because I come from the .NET world haha
Fno i just gave you answer
PThis seems like a lawless land
LI don't care how broken it is if it does what I need. I give up on TensorRT I'll come back when the option is less broken
LI spend more time fixing and trying to install broken TensorRT than the time I'd save generating images
LI should just pay someone with a working venv and TensorRT to send it 
Fi can instlal for you
Fwhat is your current level?
Fpatreon level
LIt's entry but I will upgrade if you can get me a manageble and Stable Tensor RT build that can generate 768x1344 and 1344x768. The sizes are critical.
Fplease upgrade gold i will connect your pc
Fdo it and explain you logic
Fplease open anydesk
Fmessage me
Fid
SDoes anyone have thoughts on how I can train my personal LoRA to not include the background? I know part of this is using an earlier epoch, however before/during the training process is there anything I can do to really ensure it only trains the person?
SRemove background
SI considered swapping with plain color backgrounds (ex. one red, one blue, etc) or would white be best?
SPersonally don’t see a benefit using other than white
Sgreat, i only hestitated doing this because I thought it might lead to a strong constrast between subject and background but ill give it a shot. is there a program youd recommend using to swap in the white background or just photoshop?
Fusing different backgrounds
Fit learns this
Sso do you think if i used simple backgrounds of plain colors it wouldnt learn it as much if there was variety of colors?
Fye plain backgrounds were not better
Fin my sd 1.5 test
Fbut dont know for sdxl
Sokay, ill run some tests on plain white, plain multi colors, and anything else that comes to mind and share what i find
KHey @Dr. Furkan Gözükara I have Python 3.10.9.... Should I somehow update to 3.10.11 or am I fine with 3.10.9?
F3.10.9 is fine too
KGood.
KHow do I change the subject from person to car in SOTA?
KThere?
Fyes
KGreat
KIf the aspect ratios is just "aspect ratios", why are they specified as for example [1536, 1536] instead of [1, 1]?
Fwell to understand easier
Fbut same as 1:1
Fcheck out my LLaVA captioning
FLLaVA 16 bit batch image captioning is using 27729 MB VRAM
Installed via my Auto RunPod installer
I am captioning style images for a new project for a…
Installed via my Auto RunPod installer
I am captioning style images for a new project for a…
Mmy experience as well. on first start up of a1111 I can get tensorRT to work. after that it just gets stuck on errors. It does not seem reliable yet. I also spend too much time trying to get past the errors.
Fwe solved his problem working now
୧Can anyone suggest a good anti-nsfw/adult negative prompt?
୧Not sure what to do to ensure adult stuff doesn't pop up when generating people, don't want to get banned again
Fi think depends on your base model
Fsome models are too nasty