Is Stable Diffusion WebUI Forge Just Plain Better Than A1111 WebUI?
ฝัง
- เผยแพร่เมื่อ 3 ต.ค. 2024
- Stable-Diffusion-WebUI-Forge is just plain better than Automatic1111. There. I said it :p Why? Well, performance and resource usage to start with. Even better still, some of the oldest bugs related to model loading have gone, plus you get loads of stuff just built right in! All that complete with the familiar A1111 WebUI gradio interface. Sure seems better to me!
Want to support the channel?
/ nerdyrodent
== Links ==
github.com/lll...
== More Stable Diffusion Links! ==
Faster Stable Diffusions with the LCM LoRA - • LCM LoRA = Speedy Stab...
How Can I Animate A Stable Diffusion Avatar? - • Create your own animat...
Installing Anaconda (Python) for MS Windows Beginners - • Anaconda - Python Inst...
Consistent AI Characters in ANY pose - • Reposer = Consistent S...
How To Make AI Videos - • How To Use AnimateDiff...
![ผัวเถื่อน - ยูริ โตเกียวมิวสิค [ OFFICIAL MUSIC VIDEO ]](http://i.ytimg.com/vi/P08ORRVDGl8/mqdefault.jpg)
Finally an UI for us who are not full-time unemployed in order to master ComfyUI
I think the frustration comes from tutorial youtuber keep gaslighting by saying "easier/simple"
@@Milo_Estobar i just tried forge ui and it faster
I had just installed this fork and I'd definitely be interested in seeing a video of you exploring some of these built-in features further :) Some really cool stuff in here.
haha I thought A1111 just got screwed up after using it for a while. I didn't know it was from changing models, thanks!!
Long time user of A1111 and ComfyUI, i'm truly impressed. You can run turbo models, generating 832x1024 in less than a second on 3090 but even better, you can crank up to 1640x2048, generate in less than 5 seconds and no VRAM implosion. I'm not sure if it's a good idea to change branch on my original a1111 though, i run forge on a new install, just symlinked my models and extension folders. Just wow 🎉
I can do the same on a 3060, this is definitely the new go to WebUI for me
@@xXxDisplayNamexXx don't remove a1111 yet, deforum and animatediff are not fully working in forge.
@@DrMacabre what folders did you symlink?
@@PunxTV123the whole "models", i ended up leaving a1111 extension alone as some are specific for forge like animatediff.
Holy shit. I no joke was about to go install automatic1111 since I haven't used in a year or so since I have fallen in love with instantid, but this is awesome!!! Hope it's easy to set up and use I know what I'm doing this afternoon. Thanks!
This is really nice - works way faster with my 8GB VRAM ! Thank you ! 👍👍👍
Would you say it's comparable to fooocus?
Glad to hear it!
@@NerdyRodent Hey Nerdy, how does performance compare to comfy? And do you prefer this ui over comfy atm?
While I'll definitely be giving this a go, I've never had the issues you mention with A1111 but that's me. This Forge does come with some nice features built in and the other plus side is that settings will easily transfer across, along with a couple of extensions that aren't default. Nice one.
The bugs started with mainline A1111 1.8.0 onwards. Mainline has been suffering technical debt issues since SDXL support was rushed in and has only gotten worse since. I stayed on vanilla 1.8.0 until switching to Forge, and you have to restart after switching between SD and SDXL on 1.8.0, but at least it doesn't have miserable performance on top of that like later versions.
Looks interesting, I hardly use a1111 these days, preferring comfyi but I will check this out
Your voice is so amazing , you deserve to be a movie actor
I hope Stability matrix adds this package.
I love the humor in your voice! It always gives me a good mood.
I can't believe it!!! thank you! i have a poor RTX3050 4GB VRAM, since the last A1111, i can run the sdxl but for generating a single image it was used to take about 10 minutes. Now with this release, an image will be generated in 2 minutes!! JUST WOW!! JUST THANK YOU
2 min is still too much may be sticking to comfyui
yes. depend of your hardware. maybe in your case it's better. but for my case comfyui it's pretty slow. Personally i like ComfyUi, but (for now at least) ForgeUi, gives me better performance (for my case and uses)@@anushdsouza9632
Sounds like you're generating from your CPU. I was making sd15 images with me RX570 4GB which is much older than your 3050, and it only took me 40 seconds, and that's using the DirectML version which is even slower than using CUDA.
wait.. are you talking about SDXL right? but i've now remade the calculation, it tooks 1 minute and 20 sec.. for a SDXL 1280x968 and adetailer enabled@@z1mt0n1x2
@@z1mt0n1x2How can I run Forge with Amd card?
amazing! thanks for always bring nice tips and tricks nerdy!
I was impressed how quickly the Forge variant created generations within seconds, with the A1111 version. I'd end up getting error issues with my memory even though I have 16GB of RAM & GeForce2060 Super. Usually I am so use to waiting about 2 minutes, sometimes 17 minutes as well with A1111
Even after almost 5 months with no updates. Forge is still MUCH faster than A1111. In my case for exactly the same task - it's a factor of about 5 or 6 times.
Same sampling method, images, model, weights etc 3070 RTX
4 minutes Vs 50 seconds.
It's amazing just how slow A1111 still is. 5 months and they haven't caught up at all.
I like your sense of humour, great video :-) I'm getting roughly a 50% speed boost on my RTX 3070
Regular A1111 had issues. VRam spiking and crashing the program and parts of old model retained. This Forge variant controls the memory much better, with no more spikes and no more crashes. The previous model is now completely cleared and there is indeed a speed improvement, in my case of around 20 percent, making it slightly faster than even Comfy now. I'm very happy with Forge.
interesting. what i'd like to know is what makes Forge that faster ? is there any drawback to it ? is it forced non-deterministic 'x-formers' ot things like that, or is it
definitely faster with the same parameters vs comfyui & A1111 ?
Holy crap, I've been using Easy Diffusion, because well it was an easy one click install, I figured my old 2080 Super was the reason for how slow it was, This runs circles around it. 2 to 3 minute single image generations, are now completing in like 20 to 30 seconds. And my GPU now actually sounds like it's cranking into high gear when I run it. And it doesn't absolutely hog my VRAM. This is such an upgrade.
How the hell do we keep up? LOL I'm on my 4th UI for SD in 120 days? Dizzy...🙃
Crazy, isn't it? Exciting though, no doubt!
You don't. You find the one that works for what you want, then keep using that. You only change when something comes along with a must have feature or update you don't have. That way you only change every so often.
Heh, the awesome world of state of the art software. If it makes it better, think that text generation moves even faster than image generation. A new texgen ui comes out every day, it looks like.
Learn the skills, not the software.
Just don't switch to every kew thing but rather have an interval where you update your environment, like 4 months
Thanks rodent. I will try it.
Thanks, followed the Ananconder install video first and then this and was able to install it using the full install method with no issues. I've not used Anaconda before and been using portables. Might have to do the same with ComfyUI as I'm guess it has the same pros.
WOW! ....very cool! I share your enthusiasm!
3:12 As a very savvy tech user who works on AI, writes GPU shaders as his day job, and is already very familiar with this ML ecosystem held up by sticks and glue, for the love of my time, I greatly appreciate installations that are just a few clicks - I have things to do.
It's insane how much better it is compared to A1111, I went from 1 Minute for 1 High-Res Picture to 1 Minute to 5 High-Res Pictures.
Finally an SD UI that doesn't crash every 5 minutes what a time to be alive.
ofc it doesn't works with amd? ;)
Some charitable soul to teach you how to use the A1111 repository, I've done everything changing it on the web notepad and I couldn't!
thanks for making this tutorial and bringing this new alternative to our attention. I was going thru their repo. It is very tempting. Will give it a bit more time to mature before forging ahead.
this repo is as functional as the original one, try it now.
@@AlbertoCasanova okay
@@AlbertoCasanova I need to safe hd space! Should I delete my a1111 and install Forge ?
You’re sounding better. Only a tad stuffy. So I keep hoping happy health for my rodent friend and maybe some random adventure too 😉
Achoo! Thanks 😄
@@NerdyRodent 🤣
Been using it for a few days now and I'm completely converted.
I stopped using A1111 for ComfyUI, there are some things that I go back to use it for, but I've gotten used to the speed of ComfyUI - I'll definintely give Forge a go.
I've been waiting for masks for IP Adapter.. nice find!
Been testing a bunch of UIs and Forge is the real deal, really polished.
This might be a noob question: Forge is based on A1111 - and BOTH REQUIRE CUDA graphics cards to work right? There is a CUDA substitute called ZLUDA FOR AMD GRAPHICS CARDS. HOW DO YOU GET FORGE TO WORK WITH AMD GPU AND ZLUDA? IS THIS AN OVERLY COMPLEX THING TO DO? I believe that there would be a lot of people who would be interested in a solution.
While it’s true that Linux + Nvidia is the best combo, A1111 works on AMD and even Intel CPU / GPUs too!
Well done my friend, thank you for that, I was facing like 97% and crashes on my Nvidia 1650 Ti on a Dell Laptop now I have at top 70% and runs much more faster, thank you so much
Does Forge fix the problem with Gradio's painting component being so limited you can't even erase from a mask?
I installed it and it is indeed better and faster. Only one thing I didn't realize to get:
I just wanted to try "ip-Adapter" in ControlNet, but I can't find the "ip-adapter_face_id_plus" Preprocessor because - I think - it is not updated to the last version. How can I update CN to the last version?
I tried to intstall CN also but it keeps disabled because there is CN already integrated. Any idea please?
I already saved the ip-Adapter models in the integrated CN Models folder, but the "Preprocessor" which is "ip-adapter_face_id_plus" isn't showing up.
Is it possible to download it maybe then upload it in a specific folder inside WebUI Forge folder?
I'm having the same issue. I hadn't thought of it not being updated but I think my Forge ControlNet version is higher than my A1111 version - I'll have to double check as they aren't the exact same extension now.
I was assuming the preprocessor wasn't being picked up because it was in the wrong folder but putting it in built-in ControlNet folders didn't solve anything. Please shout up if you find a solution and I'll do the same. 🙂
BTW, there are downloads for it under one of Sebastian Kamph's latest vids.
@@Elwaves2925 having the same issue!
@@saranmt1 It's listed as a high priority bug on the github, so there's nothing we ca do for now. It may already be fixed if I can understand github correctly but it's not released yet. Keep checking there for an update. 🙂
u need to change .bin to .pth
Love it, can finally run SXDL in webUI with my 8GB card.. and agreed no blasted memory issues
What about Nsfw ?
Well, the new Lora Tab make it completely unusable for me : The loss of the "folder" button for a cumbersome "tree view" is... Why , Just why ?
The entitled pessimist me is whining for realizing I've done everything wrong from the start and have to uninstall everything (Auto1111, Ooba, ST, the works) and start over and learn how to use Python environments and maybe even learn how to code while I'm at it and so much hassle just to generate images, while the adult me is slapping pessimist me and telling him to quit whining and get to work, there's nothing wrong with learning something new and useful. Oh and I hit sub. Thanks Nerdy, hope you're feeling better.
On my tests (8gb 3070) I get images almost twice as fast as with normal a1111, but one thing confuses me: Everything the same I still get different images than in a1111, even if the github says there shouldnt be a difference.
are u sure u using the same settings and seed?
Nice! thanks for the video! Does the A1111 maintainer not merge Pull Requests regarding the issues hence someone decided to fork it and fix the base?
couldn't go back to A1111 after it seemed to constantly break itself during updates, and confyUI was just way quicker, but way more annoying to use. Forge is a godsend
Just upgraded to the pro version of Forge totally worth it what a beast.
But...is it good for AMD gpu? Please tell me it's good so i don't have to buy another card
Downloaded it last night(17/02/24), this one is going to replace ALL my other tools, I'm running a modest setup and this thing is blazingly fast!!(20 images in approximately 10 minutes)
Nice! from lllyasviel auto INSTALL!
That fooocus to Forge!
Hm, Photomaker doesn't seem to work at all. How did you get it to run?
Cracking video. Thanks!
Glad you enjoyed it
Excellent! Thank you.
Great overview video!! I am trying to figure out where to place upscale models... do you know folder location?
THX for this, much appreciated ! I just can't get use to the spaghetti comfy-ui even if it's much better so for me this is sweet music. I have tested it a bit and surely, it's faster (running on a 4090) and the ability to use masking in controlnet is fantastic. Thanks and I hope you post some more on WebUI Forge...
waiting for the comfyui node of whatever this is doing
Forge is very painless! thanks for the recomendation
does the fun ever really "stop there" with NeRo? I think not.
Forge is Blindingly fast. Before I used to make one dozen 512x512 images would take me 1 minute. Now I make 768x768 and 1024x1024 images under 5 seconds with the newer LCM models.
Img2Img doesn't really work for me with Forge. It always spits out a mess or a lower quality.
your mom as a cute rodent had me laughing
I've installed it today and it is faster than A1111 on my RTX 3070. It uses less V-RAM.
So much better for low-mid tier PCs. Easy install too.
loras dont load up.....
what about teaching followers how to setup a docker container for these OSS sources? I always find it scary that people run third party code without a shred of fear, at least docker puts everything in a sandbox
I got into automatic1111 because of this video, but gave up forge for the original. Forge has lots of issues I dont see with the original
I personally never had problems with changing models in A1111, is it only with SDXL models?
I haven't either and that's with 1.5 and SDXL models. I constantly switch between Realistic Vision and RealVisXL while keeping the same prompt.
x2
neither did i but i understand what he's saying with that with each change it didnt release your ram, thereby stacking up and making things slower. I watch vids on my second screen and they stutter on my 4070ti while i generate images on A1111, this isn't an issue with comfyui or now Forge for me which is much more user friendly than comfyui and was the main reason i still used A1111..
After switching to Forge UI, I can't activate "ReActor". Is this still an issue? Any leads fellas?
A .sh file is a "shell script", that's what you call it
Made a joke about the dot SsHHa earlier and then my comment got deleted lol. TBH, I found it amusing and funny in a good way about how he pronounced it throughout the video.
He wants us "to learn" and not just do, he can learn from us too. Unintentional rhyme @@BoyanOrion
What about AnimateDiff + Controlnet via Forge Webui?
animatediff doesnt work, along with a boat load of other extensions 😔
Back to a1111 and leave the noodle insanity behind thank God.
Thanks for this! Like to try out everything.
when changing model cmd crash. i have to lunch it every time i change model. any help?
Can't get LORA's to load with Forge... anyone have this issue?
The question for me is does Forge work for AMD GPU?
If not, then it's not better to me.
Neat, well I can't even get as far as downloading it because the file seems to be corrupt lol.
Did you use 7Zip to unpack it?
webui-forge and a1111 are great ways to discover more errors than you ever thought were possible, IF you can manage to even install them. and to make matters even more interesting, A1111 is being maintained and webui-forge is not. I've got a 12GB card and it's not enough even though their troubleshooting wiki has examples for 4GB card users. --lowvram nor --medvram worked, --xformers will never be installed, ipadapter v2 will never install, etc. Changing checkpoints will freeze the whole computer, just loads of learning experiences. No idea what they're thinking.
... a cross between auto1111 and something that works! ha!
... But is it better than comfy?
What's the update? Is this still supported or updated? or is it dead re updates?
1060 6GB VRAM (same prompt and image size 512x768) - Auto1111 [A: 2.81GB] [R: 3.37 GB] [Sys: 4.4/6GB] Time 34.5 sec WebUI Forge [A: 3.63GB] [R: 4.13GB] [Sys: 5.4/6GB] Time: 32.8 sec
I wanted to switch to ComfyUi. Do you still recomend using comfy when forge is released? (I have A1111 skill)
I use both. I would recommend diving into Comfy Ui because you can learn a lot about backend settings and control over every aspect of AI image rendering
@@PaulByrd1 and what about speed? Thanks
i can't load SDXL, When i tried to load SDXL it's says just " Press any key to continue . .."
Actually huge thanks to video maker, Installed this webui Forge and it is indeed AMAZING!!
Simply marvelous!
Interesting, hopefully all those changes get implemented upstream by A1111. I don't see a reason for the codebase to be split if there aren't breaking changes.
Can't run deforum or mov2mov though. So in generating images faster it is.
''your mom as a cute rodent'' got me 😭😭
Oops... update.bat calls environmental.bat which gives an error. 🙄 Ah... fixed it. VPN security blocked the bat. Running now, and wow, it is faster than Fooocus or Automatic 1111! Intel i-7 and MSI 3060 OC 12Gig and 32Gig RAM.
How does this compare with vladmandic's SD.Next? Is it worth switching over to Forge?
How to run instant-ID in forge?????
Just tried your model switching example with A1111 and it worked just fine. Maybe patched or intermittent?
I can't get the t2iadapter controlnet for style to work in this. I don't have the clipvision preprocessor and it's not automatically downloaded when I try to use it.
This thing lets me run SDXL on hires fix 2x without my machine even blinking!
Cannot get animatedif working with Forge
My Forge is missing the IP-Adapter models, the list is empty, and thus, the function doesn't work. Any idea what this could be, or how to fix it?
Doesnt work for me. I get an error saying torch is not able to use gpu
What's the simplest way to create animated video from an image? I want eye blinking, but Pixverse is too warpy, random and uncontrollable.
why they do not create a pull request ? This going to create a division and lose many updates and new feature from the original tool.
Help please, why Lore don`y working at this version, it`s generate fast and good, but Lore don`t working, in version which crashed but rarely gif me art all Lore was working
Hey NR, do you scripts your videos? Or work in sections and edit as you go?