Downgrading A1111 to older versions to solver some extension problems
ฝัง
- เผยแพร่เมื่อ 1 ก.ย. 2023
- This video explains how to downgrade A1111 to older versions
to upgrade again, just use git pull
alternatively, you can try deleting VENV folder, this may solve some extension issues but not all of them.
downgrading maybe necessary for some people who rely on some extensions in their work and this matters more than the new updates.
in newer A1111 1.6, some extensions such as Mov2Mov, Deforum and several others didnt work for me thus far, hopefully updates in the future fixes this... but this is not the first time A1111 newer versions makes older extensions stop working, so knowing how to downgrade could be a useful piece of info.
NOTE:
I personally have two folders, one for 1.5.2 version for instance and one for 1.6 since 1.5.2 seems more stable and use set COMMANDLINE_ARGS= --ckpt-dir "E:\stable-diffusion-webui\models\Stable-diffusion"
along with other (--xformers ---update-check --no-half-vae--medvram-sdxl )
so both refer to the same models to save disk space. - วิทยาศาสตร์และเทคโนโลยี
1.7 just came out and this tutorial help a lot to downgrade back to 1.6. Thank you so much
Glad it was useful, You are welcome
I was searching for days how to downgrade to 1.6.0. No one knew. THANK YOU for this easy expert solution! 🙏🙏🙏
I really appreciate you for making this video. It solved my issue. Thank you so much.
not sure, but i think its the same file on mac but named webui-user.sh and you edit it in the same way and put the command there, if you go throught he .sh file it might give you some guidance
Thanks SO MUCH - you saved my time!!
Thanks for your tutorial
You're welcome
THANK YOU LOTS!! it worked
thank a lot
Yes true, this is the issue of version 1.6. For my practice, keep my older version in one folder, and create another new folder for just testing ver 1.6.
true, I use --ckpt-dir "E:\stable-diffusion-webui\models\Stable-diffusion" command in Command Args in the webui-user file to refer to the other folder models... hopefully there will be no need soon.
I have a question: What are the correct commands if I also want to change the references for Loras and CN-models (which are in the extensions-folder) or even the complete extensions folder ?
unfortunately, in A1111, I am not sure if this option is available for LoRAs....in Comfy you can refer to the models/lora/controlnets from A1111 directory , but in A1111 only path to Check point is part of the command line args as you said earlier.... you can test for Controlnet in A1111 (Settings, Controlnet, Extra path to scan for ControlNet models) but i have not tested it thus far.
Will this prevent any auto update to 1.6 in the future?
no, i copy A1111 1.6 folder (without the models) and downgrade it to 1.5.2 to use some older extensions... and keep another with most updated version ... and use --ckpt-dir "E:\stable-diffusion-webui\models\Stable-diffusion" to refer to the same models directory, so i dont need to copy the models... any way, you can always just use git pull and get updated again to 1.6
I downgrade back to 1.5....but generate doesn't working, it wait for few seconds and nothing happened
if you had problems with downgrade, you can delete venv folder inside the \stable-diffusion-webui folder and re run again (THIS will install the suitable libraries again and will take a while, bu this might help any conflics you get), hopefully that works for you.
You should also consider moving to ComfyUI, it is more reliable, I wanted to create Videos about Comfy but couldnt find time for that, hopefully in the future I will do, A1111 is having too many bugs recently, it is not going so great.
@@AI-HowTo does upgrade the web UI will affect the result of generation? will that have bug on running checkpoint or lora as well? I don’t know may I overthinking or not … I found v1.7 is not as good as v1.5 regarding the gen result.. does upgrade the webUI will also upgrade the stable diffusion AI itself ?
no, using same seed, same settings, should generate same output despite which A1111 version you are using, unless some bugs are causing LoRAs to remain stuck in memory for example...so when you remove the LoRA sometimes, it is still in memory and affects your generation without knowing it.
I see bro, thanks, I didn't change any setting and remove any LORA, still using the same LORA as before, maybe it's just my too overthinking lol....@@AI-HowTo
I run 3 different versions, one dedicated to 1.5, another to XL and an all around one.... I upgrade the XL model continuously, but find that as you say.... somethings not upgraded to upgrade. I mainly did it because all the models of XL fit on a 2T drive.... I need a 4T drive for my 1.5 models and another 2T for all around, oh and 2 different Kohya ss locations as well, for XL and 1.5, Once the Titan comes out or I buy a 6000, I can put the 4090 on a machine just for modeling and LORAs.......what a mess
indeed very messy, its is better to have different installations as you mentioned, and better to refer to the same model directory using COMMANDLINE_ARGS --ckpt-dir option ... one installation simply doesnt work for all use cases, a common problem in community open source free models...hopefully this is improved soon....to run optimally indeed you need good GPU/and lots of disk space.
I remember the redirect to models that used-to-be, it vanished. So I know its dumb to have multiple same models installed on diff drives, but I figure....it's like a backup. Putting together a 8tRaid to put XL on and redirect others to models folders... @@AI-HowTo
I must quote another genial solution (for people like me who are users and not programmers) which I found below in one of your answers:
" I use --ckpt-dir "E:\stable-diffusion-webui\models\Stable-diffusion" to refer to the same models directory, so i dont need to copy the models"
Next problem solved, saving hundreds of gigabytes of space on my SSD!
Thank you !!!
For 1.6 I had to disable most of my extensions, I know for sure Roop was causing a conflict. 👍
yes, some extensions no longer work properly, so I had to run another copy downgraded to 1.5.2 and one with 1.6 to get all extensions working properly, hopefully extensions are updated soon, otherwise, having another A1111 copy will become necessary to get all things running smoothly at the same time.
@@AI-HowTo I've noticed since going to 1.6, think I will probably go back to 1.5 until things get smoothed out. 👍
it still doesn't appear for me, I'm going to restart everything from scratch, do you have a guide from scratch for the process?
these tools can be clumsy it appears, since they dont work the same for all people.... in th-cam.com/video/RtjDswbSEEY/w-d-xo.html in the first two minutes it shows the installation steps I think you can skip downloading python or git sinc you have them already and just git clone github.com/AUTOMATIC1111/stable-diffusion-webui.git then run the webui-user.bat in a new folder for instance, I dont think they changed on just check the A1111 for a fresh installation , this can be good sometimes when dependencies are messed up which results in some conflicts between various extensions.
Someone know the tag of the version v1.6.0-2-g4afaaf8a?
The v1.7 is totally a disaster, I wanna back to 1.5
Didnt work for me. I updated to 1.7 then tried to revert to 1.6. After hard reset my SD couldnt find Python . After reinstalling python it gave another error. Delete venv folder also didnt help. I have to fixed it manually by putting new line in requirement_version. But then loading models gave error. So I have to update to 1.7 again and it solved every problems.
I see, A1111 gets more buggy everyday.
how did you solve couldnt find Python roblem
this version dont work on amd r they direct ml or exclusive for nvidia
no, as far as i know it is not, downgrading is a python and github thing, not related to anything else.
@@AI-HowTo not all version work with amd as far i got but i able to downgrade by trying out multiple old version lol
good, trying different versions is a good way to reach to a solution with these messy open source projects
how do you do this on a mac? .bat is for windows correct?
not sure, but i think it is the same file but with .sh extension, and git reset is a command that is independent of the operating system.
I tried this but It didnt solve this problem I'm having, since I made a clean installation, I can't use lora anymore. I have a long list of errors for each lora i have in my folder
sometimes deleting the venv folder inisde A1111 installation could help solver lots of errors, some A1111 are buggier than others.