Been a Forge user for a long while now and it's worth it for the speed increases over A1111. There was some initial problems with some extensions not working but I found a fork for them tha was Forge compatible. Thankfully, I don't have too many extensions. Once that was sorted I ditched A1111 straight away. Forge continuing to be be updated was great news and the addition of Flux was even better, especially as loras are working with it now.
I'm still waiting for them to support composable-lora and two-shot after a year. Those are pretty crucial for composition with multiple loras. If they could do that or have some reliable compatibility mode, I wouldn't need A1111 anymore
@@steve_jabz What's two-shot? Is it something like regional prompter. When I first switched to Forge a good while back, extensions was the one thing that stopped me fully switching. Then I found forks that offered Forge versions and I've never gone back. Have you tried looking for Forge forks, I had to dig quite far and there is an element of trial and error.
@@SupremacyGamesYT The fact is Forge makes A1111 obsolete, and since it's build on top of A1111 it's simply a natural evolution, you're missing the big picture. But BY ALL MEANS continue to use A1111 even though Forge is available just so you can feel better
@@Lexie-bq1kk Forge does NOT make Automatic1111 obsolete, Forge is just another tool, in actual fact Forge has some issues that Automatic1111 does not, especially with things like extensions especially ReActor, you need to understand that the latest is not always the best for what people need, don't be a control freak.
I don't even feel like flux is that big of a deal right now, unless you really need text generation in your images. There's so much community work that still needs to be pushed into the models and optimization until I'll be really impressed, but that's just me.
shifted to ForgeUI already, working with fluxDev nf8 on my laptop with RTX 3060 6 gigs of vram and 16 gigs of ddr5 ram, it takes around 3 minutes to generate an image, it takes long but the results are great.
It's amazing how much more I understand about the workings of workflows since switching from A1111 to ComfyUI. All of these Forge settings make so much more sense to me with that Comfy knowledge than I ever understood with A1111.
Moved from A1111 to Forge about month ago and very satisfied, highly recommend to set Forge while keeping Automatic too. Speed is much higher like 1.5 times for my scenarios with SDXL/Pony, and it even doesn't overflow 32gb ram like A1111 did. Note for those who will use Flux with GPU like mine 3070 8Gb (for other memory sizes not sure) to have at least 64Gb ram, in that case it will not use swap and kill lifetime of SSD. In the beginning I had 32Gb ram and it worked, but I noticed it used up to 10Gb of disk space at C drive quite often with image generations, which is quite a lot and for one session it could be more than 100Gb easily, so I decided to upgrade up to 64Gb to be on safe side with SSD lifetime resource (and generally be faster). About extensions, Regional Prompter doesn't work for sure (but instead you could use SD Forge Attention Couple). Works fine: ADetailer, Dynamic Prompts, ReActor, Civitai Helper, ar-plusplus, Photopea.
You have somehow failed to mention the main reason for Forge existing in the first place: It is much better at memory management than A1111. Especially with low VRAM cards, it'll run much faster and achieve higher maximum resolutions. The main disadvantage is that quite a few extensions don't work. Unfortunately, the regional prompter extension is one of those.
Regional prompter works. I just installed it the other day as I'd completely forgotten that it existed. Although, to be accurate this was in a pre-Flux version of Forge that I use solely for SDXL and SD1.5, so that may make a difference.
@@Steamrick Ah okay, thanks for clarifying. I figured that might be an issue and that's why I mentioned the pre-Flux version. It's part of the reason I have multiple versions of the UIs. 🙂
It doesn't have a better memory management than A1111, it loads and unloads loras for every single image, it's hell on earth if you work with a lot of loras.
Olivio, thanks for another great video! I'm a long-time A1111 user trying to get a handle on using Flux with Forge (because Flux seems to be the way everything is going now). The biggest thing I miss in Forge are some of my extensions. How can I tell which extensions work with Forge as well as A1111? The biggest one I need is a save-state extension so I can save the entirety of my settings and come back later and start where I left off, instead of trying to remember what exactly I had setup before. Any tips in that regard? Thanks!
A1111 is still great for SD1.5 and SDXL, I got a ton of extensions installed on it and I have no intention of installing all of them again on Forge (with the exception of Ultimate SD Upscale), I'm currently using Forge mainly for Flux only.
@kiransurwade3576 You mean "control net" ...hmmm, I think control net for Flux is still a Comfy UI only thing so far (last time I checked, the tech develops so fast it's hard to keep up). A for LORAs, yes Forge does support LORAs for Flux.
To be true A1111 is a solid pro classic UI. Dozens of people are working on it. And yes, it doesn't support Flux right now. Maybe waiting for something... E.g. for the Pony v7 and Auraflow idk.
Where have you been!!!?? LOL I've been hunting and pecking for two weeks to put together the pieces. And here you lay it at my feet? Harumph! Love ya! OH! Oh! GGUF? Next video???
IMO, it's way easier to setup Forge, Fooocus, SwarmUI, ComfyUI, etc via Stability Matrix. One click install, models can be downloaded via Stability Matrix and shared with the other WebUIs, image output folders can be shared, too and you always get the most up to date version of these WebUIs...at least for Forge, Foocus, and ComfyUI, which I have installed on my PC.
While I really like ComfyUI as I like building out the nodes and seeing how it all works and how changing them around can affect the output, I can't get the new Flux models to work without needing insane render times if it would even render as I constantly get cuda out of memory errors (this is on a some what powerful machine with a 3080). That also meant I couldn't use any of the stuff I would want to like IP adapter or Controlnets. That and Comfy seems to keep giving me more and more errors for whatever reason so I decided to try this UI and I was amazed at the difference in render times AND I didn't have to use the gguf models which also gave me issues in Comfy so yeah, if you are having issues even trying to play around with Flux models, give this UI a try.
Question. i know about Forge and comfy UI. i know then comfy is more complicated but I wonder how it is about speed and memory. What will be better to use with low VRAM GPU, like 3080 10GB, 32GB RAM, 5900x, Win 10? I want to start with ComfyUI but i will start to learn next week. But it would be cool to know if one is more effective then others.
Invoke got a nice update lately tho. And it's like on the another end of the UI/UX principal to the ComfyUI. It's solid and easy to use. Artist friendly. Also offer some level of node workflow which isn't as powerful as Comfy but instead make someone that knows the nodes could make an automated tool out of it for a non-node user. Which the interface also offer to hide the node and see only the selected inputs and the user just press the generate button to run that workflow.
Does the local/community version have feature parity with the priced options? I remember seeing Invoke a year or so ago and was really impressed with its ux/ui. The AI scene is just f--king exploding constantly and I want just one tool I can really settle into and learn really well. I know ComfyUI is probably the most robust toolset but its ux/ui looks like a nightmare. At that point I'd rather just paint the actual images mysef.
Great video! I'm a long time A1111 user, and I just installed forge 2 weeks ago. It's really, really good, but It's not perfect. It seems to have some UI issues, you can't use the refiner currently, and it maybe has some issues handling memory on occasion. But it can use Flux models with ease, and it seems to render and upscale faster than A1111 for me at least. I'm keeping both.
Thank you! I just did this, but didn't select any vae or text encoders and it generated fine. Maybe those are needed only for normal dev? Seems like they're integrated in the nf4 v2?
I installed Forge a week or so back. I never have any luck installing as snag free as youtubers do! I do recall I had to upgrade my CUDA which was straightforward but be careful to upgrade to the level of the forge install you are using. Its a good idea to carefully watch the execution window for clues as to whats gone wrong (often models don't download completely and you have to manually do the download again) If you keep getting errors restart forge and then copy the whole dialog from the window into perplexity or gemini to get advice.
Its not a waste of time to still be using A1111. After trying both Im still using A1111, forge might have been slightly faster for each image (less than 10 seconds) but it comes with a bunch of extensions i don't use and doesn't support others that i do. Also A1111 hasn't broken on me yet wile my forge stopped opening out of no where so im not sure whats up with that
Switched to Forge recently. Works nice, just a bit disappointed with the Flux model. And I'm still not able to activate Clip Skip for XL/PONY. Or is it this Clip Skip for SDXL checkbox? Anyway. If you don't want to play around with Flux, you can still use A1111.
I've been using for months.... there are some downsides with some plugins, but usually works pretty well. It renders SDXL LCM models in less than 10 seconds on my 4GB VRAM
@@eduMachado83 what gou though? I can gen sdxl on my 980ti 6GB but it’s more like close to a minute (which is wild because originally it took like 6 minutes in comfy)
Forge has moved to beta forever and we be a testing ground for features that's going to be ported to A1111 after testing. So it will always be ahead of A1111 in terms of commits but A1111 will always be more stable
Just a heads up for people having extension issues in forge, if forge gets stuck at 50% when generating its because your dimensions are not divisible by both 8 and 32. In forge this is common because the slider goes in 8 by 8 increments. All numbers divisible by 32 are divisible by 8 but not every number divisible by 8 is divisible by 32, that's the problem. If you want to avoid the hassle of calculating, under the config file change it so the increments of the dimensions sliders are 32 by 32 and you won't have that problem just be careful with the upscale make sure the result is also divisible by 32. This is useful specially if you use regional prompter and other extensions, they won't crash anymore For my needs most extensions work, except for composable lora advanced functions but i honestly don't need it.
I think I'll learn Comfui just for Flux. Currently the generation speed on Flux for upscaling is significantly better compared to Forge. Upscaling is essential for me.
am getting the error message: "connection errored out" when I try to generate a picture or do anything. Anyone know how to solve this? I've got an rtx 4090 and 32gb ram, it should be enough no?
I just switch to Forge after spending almost 2 years on A1111 , took almost no time to set it back My laptop with 4gb GPU couldnt handle SD.XL on A1111 but it can on Forge , its a little bit slower than SD1.5 but it give me way better results. i definitly dont regret installing Forge
Forge still doesn't support flux lora. When i use flux models, forge keeps moving the model between my RAM and vram, before every generation, despite that i have 16gb vram. Which makes it super slow.
It supports lora. Include nf4. Just need to activate the fp16 lora from the top menu. Then add your lora. Annd yes its slow then comfy. 38seconds vs 1 minute with my 4060..
Err, yes it does support Flux lora. As well as fp16 stated above, you may also need to drop the GPU weights value down (in the 8000's is a figure I've seen suggested). It isn't quick but it definitely works because I'm using them with 12Gb VRAM. Slow doesn't mean not working.
@@UgurDoydukFM bummer, i get FASTER RESULTS with FORGE on my 6GB rtx 2060 using the regular fp8 DEV model. At the suggested 896x1152 i can do 20 steps in 1:30 seconds! and a 1080x1440 image in 2:30 mionutes!! how come you have a BETTER gpu and almost the same as me?
Forge is now a funnel to test features for A1111. I don't know why you don't mention this. A1111 is the way to go if you don't want to be a tester of new features that might/might not work. This video is a good reminder as to why I left the toxic and misleading community Olivio runs.
You know you could script up and distribute docker compose files with a similar amount of effort as editing one of your batch files? After a one-time setup of WSL2, Docker Desktop (free), and NVidia's Container Toolkit, people could run your script to download everything into a ready-to-go container (like a lightweight VM) that will allow them to run the stuff while also insulating them from the possibility of malicious software. Software like this is a perfect fit for containers.
Sounds cool, but i have literally no idea what you just said. So it's cool that that works, but you need to know how. And at this moment in time i don't
For me it's too rigid. While comfyui is far from perfect, it's a lot more flexible and allows you to pick up quickly where you left off various projects.
I replaced a1111 and forge with Krita ai extension for convenience and control. It runs on comfyUi but it's much less complicated for image control and tweak.
I think ComfyUI has a really steep learning curve to start out with, but once you understand it you're good. Good to know there's an actively developed successor to A1111 though.
I don't know why it didn't get mentioned or I missed it - Forge is mostly a downstream version of A1111 webui ("Forge is a platform on top of Stable Diffusion WebUI") with stuff added to it, so it wouldn't probably wouldn't even exist without the work put into the original webui. Also, there is no point of moving or copying large model files, better create one big folder where you store the models, then create symbolic links into A1111 and Forge folders. That works nicely when you have 5+ different UIs for SD, so you don't have to move models around and you don't have to bloat your installs.
Should I use Comfyui? A1111? Forge? I think I watched the exact same video but instead of forge it's comfyui. Which not gonna lie, I still like comfyui way more than A1111. But is there a better UI already than comfyui? Honestly, the only complaint I have with comfyui is that it's quite unstable. But it does a really good job of experimenting with new ways of generating ai art.
noob here. do i need to worry about git when having both a1111 and forge on my machine? will i get any git directory errors? would it be wise to uninstal any and all software related to a1111 before changing over to forge'?
Is it normal that it takes forever to extract files after downloading One Click package? Iv been sitting for over 40 min and files er still extracting...
Using it atm because of Flux, BUT there are clear downsides: It can't do proper hashes for cross-posting on Civitai (Have to manually replace the garbage hashes with the proper ones in each image and I frickin hate it). Several extensions, that work in A1111 don't work in Forge, so yeah ... cool it a little with the hype, maybe?
Good video I like your videos. Unfortanly, I had some less positive experiences after that; they make it flux-compatible I have regular black images, while the version below it work very stable, be a ware I am running on mac m1
Hello, I have 35 folders in my models in ComfyUI and there is only 6 paths in your script, is it normal ? A lot are missing like unet, ipadapter, inpaint, etc…
At least in regards Colab Notebook versions of Forge--there is still no support for Controlnet batch processing. Forge only processes the first image. Been an unresoved bug going on seven months.
set COMMANDLINE_ARGS= --no-half --lowvram --opt-split-attention --xformers --api currently using this IN A1111 and do i need to use lowvram on forgeUI? 4gb VRAM user here
Forge UI was abandoned, and then restarted with flat out comments by the author that he was building it for himself and not to expect it to be useful for production use. For that reason alone I'm very hesitant to use it. A1111 has its issues, but it hasn't been abandoned. I do use it for Flux, but that isn't my primary model right now.
I switched from A1111 WebUI to Forge WebUI because Forge performance is faster than A1111 with RTX 3050 8GB GPU when running SD1.5, SDXL, and PDXL models. I ran the Flux 1 dev model with Lora takes 1-3 minutes for 1-2 images. If without Lora it can take 30-60 seconds
@@jimg9820 While it may be harder to pick up initially, when you do pick it up it is genuinely unfathomably more convenient and you’re able to do so much more than a gradio ui, that came out what, 4 years ago now?
I tried forge for FLUX but it seems unstable on my computer, while SwarmUI works fine. The only issue with SwarmUI is that it doesnt save the image metadata in the images themselves.
I had to change my slashes for the directory from backslash \ to forward slash / I've read in some cases it needs to be double [//] and your a1111 install location may be different so do not copy his exactly.
@@Jakeurb8ty82 I put the double[//], now its opening ForgeUI, but models are not available, I put the right address, in my case, C://pinokio//api//automatic1111.git
Tried this and then uninstalled it not long after, I like to make img2img animations and guess what? img2img batching is completely broken on forge, all it does is just generate the first img of a batch over and over and over again, and the developer doesnt look like they have any plans to fix it anytime soon. To save you all the trouble, stick to regular a1111.
Good day sir! I'm very new to all this SD thing. One issue I encounter is that when I create images with my favorite model, it only works well at a specific resolution. When I increase the resolution, I face the problem of duplicated characters. Is there any way to solve this issue? How can I create an image with my favorite model at the desired resolution? Thank you for taking the time to respond!
Not the OP, but check if the UI has hires fix. Generate the images at the resolution they're most ideally generated from, but turn on hires fix and change the scale to upscale after the first generation.
Been a Forge user for a long while now and it's worth it for the speed increases over A1111. There was some initial problems with some extensions not working but I found a fork for them tha was Forge compatible. Thankfully, I don't have too many extensions. Once that was sorted I ditched A1111 straight away. Forge continuing to be be updated was great news and the addition of Flux was even better, especially as loras are working with it now.
there is big a problem! many extensions don't work with this ui.
EXACTLY !
I'm still waiting for them to support composable-lora and two-shot after a year. Those are pretty crucial for composition with multiple loras.
If they could do that or have some reliable compatibility mode, I wouldn't need A1111 anymore
@@steve_jabz What's two-shot? Is it something like regional prompter.
When I first switched to Forge a good while back, extensions was the one thing that stopped me fully switching. Then I found forks that offered Forge versions and I've never gone back. Have you tried looking for Forge forks, I had to dig quite far and there is an element of trial and error.
Yeah, if that's the case, it's not important. And stopped development.
Unfortunately thats true.
And A1111 is a solid solution, an idea and a help for so many people... How could one critic that?!
classic clickbait
Hey, some people use old phones instead of smartphones. Just be aware you are using outdated stuff and becoming a minority.
@@SupremacyGamesYT The fact is Forge makes A1111 obsolete, and since it's build on top of A1111 it's simply a natural evolution, you're missing the big picture. But BY ALL MEANS continue to use A1111 even though Forge is available just so you can feel better
@@Lexie-bq1kk Forge does NOT make Automatic1111 obsolete, Forge is just another tool, in actual fact Forge has some issues that Automatic1111 does not, especially with things like extensions especially ReActor, you need to understand that the latest is not always the best for what people need, don't be a control freak.
A1111 Is the best solution, you can take.
NOTE: IN FORGE UI, CONTROL NET STILL WON'T WORK FOR FLUX.
Why he dosnt say that in the video? 😮
@@AntonioMaciaMartinez because you can still use FLUX and many other things/
I don't even feel like flux is that big of a deal right now, unless you really need text generation in your images. There's so much community work that still needs to be pushed into the models and optimization until I'll be really impressed, but that's just me.
@@Exaltar so as of now, what SDXL model you think is better or as flux?
@@Exaltar Prompt adherence is much better than SDXL across the board, text or no text.
shifted to ForgeUI already, working with fluxDev nf8 on my laptop with RTX 3060 6 gigs of vram and 16 gigs of ddr5 ram, it takes around 3 minutes to generate an image, it takes long but the results are great.
It's amazing how much more I understand about the workings of workflows since switching from A1111 to ComfyUI. All of these Forge settings make so much more sense to me with that Comfy knowledge than I ever understood with A1111.
right? an experimenting in comfyui brings up totally new combinations and solutions too :)
Moved from A1111 to Forge about month ago and very satisfied, highly recommend to set Forge while keeping Automatic too.
Speed is much higher like 1.5 times for my scenarios with SDXL/Pony, and it even doesn't overflow 32gb ram like A1111 did.
Note for those who will use Flux with GPU like mine 3070 8Gb (for other memory sizes not sure) to have at least 64Gb ram, in that case it will not use swap and kill lifetime of SSD. In the beginning I had 32Gb ram and it worked, but I noticed it used up to 10Gb of disk space at C drive quite often with image generations, which is quite a lot and for one session it could be more than 100Gb easily, so I decided to upgrade up to 64Gb to be on safe side with SSD lifetime resource (and generally be faster).
About extensions, Regional Prompter doesn't work for sure (but instead you could use SD Forge Attention Couple).
Works fine: ADetailer, Dynamic Prompts, ReActor, Civitai Helper, ar-plusplus, Photopea.
You have somehow failed to mention the main reason for Forge existing in the first place: It is much better at memory management than A1111. Especially with low VRAM cards, it'll run much faster and achieve higher maximum resolutions.
The main disadvantage is that quite a few extensions don't work. Unfortunately, the regional prompter extension is one of those.
Regional prompter works. I just installed it the other day as I'd completely forgotten that it existed. Although, to be accurate this was in a pre-Flux version of Forge that I use solely for SDXL and SD1.5, so that may make a difference.
@@Elwaves2925 It worked pre-Flux, yes. One of the changes with the implementation of Flux broke it.
@@Steamrick Ah okay, thanks for clarifying. I figured that might be an issue and that's why I mentioned the pre-Flux version. It's part of the reason I have multiple versions of the UIs. 🙂
It doesn't have a better memory management than A1111, it loads and unloads loras for every single image, it's hell on earth if you work with a lot of loras.
Olivio, thanks for another great video! I'm a long-time A1111 user trying to get a handle on using Flux with Forge (because Flux seems to be the way everything is going now). The biggest thing I miss in Forge are some of my extensions. How can I tell which extensions work with Forge as well as A1111? The biggest one I need is a save-state extension so I can save the entirety of my settings and come back later and start where I left off, instead of trying to remember what exactly I had setup before. Any tips in that regard? Thanks!
A1111 is still great for SD1.5 and SDXL, I got a ton of extensions installed on it and I have no intention of installing all of them again on Forge (with the exception of Ultimate SD Upscale), I'm currently using Forge mainly for Flux only.
You can just copy your extensions folder I think, but some of them may not be compatible
EXACTLY! Went thru too many issues with crashes rendering my Forge unusable. Went back to A1111 and so far very stable with extensions.
Does the Forge for flux support "Control net" and "LORA" ?
@@kiransurwade3576 Not sure about controlnet, as I'm waiting for for everything to be available. Loras are supported though.
@kiransurwade3576 You mean "control net" ...hmmm, I think control net for Flux is still a Comfy UI only thing so far (last time I checked, the tech develops so fast it's hard to keep up).
A for LORAs, yes Forge does support LORAs for Flux.
Love you videos and details that you provide! Thank you so much!
Oh I remember the days so long long ago when A1111 was the new kid on the block and THA BOMB!... how times change.
To be true A1111 is a solid pro classic UI. Dozens of people are working on it. And yes, it doesn't support Flux right now. Maybe waiting for something... E.g. for the Pony v7 and Auraflow idk.
a1111 waits till they can add a future properly, Forge has a new update every day and still is not working with Flux.
@@deadlyrobot5179flux worked fine for me 🤔
Where have you been!!!?? LOL I've been hunting and pecking for two weeks to put together the pieces. And here you lay it at my feet? Harumph! Love ya! OH! Oh! GGUF? Next video???
you got a harumph out of that guy!
IMO, it's way easier to setup Forge, Fooocus, SwarmUI, ComfyUI, etc via Stability Matrix. One click install, models can be downloaded via Stability Matrix and shared with the other WebUIs, image output folders can be shared, too and you always get the most up to date version of these WebUIs...at least for Forge, Foocus, and ComfyUI, which I have installed on my PC.
wait till you try installing roop in comfyui
While I really like ComfyUI as I like building out the nodes and seeing how it all works and how changing them around can affect the output, I can't get the new Flux models to work without needing insane render times if it would even render as I constantly get cuda out of memory errors (this is on a some what powerful machine with a 3080). That also meant I couldn't use any of the stuff I would want to like IP adapter or Controlnets. That and Comfy seems to keep giving me more and more errors for whatever reason so I decided to try this UI and I was amazed at the difference in render times AND I didn't have to use the gguf models which also gave me issues in Comfy so yeah, if you are having issues even trying to play around with Flux models, give this UI a try.
Question. i know about Forge and comfy UI. i know then comfy is more complicated but I wonder how it is about speed and memory. What will be better to use with low VRAM GPU, like 3080 10GB, 32GB RAM, 5900x, Win 10? I want to start with ComfyUI but i will start to learn next week. But it would be cool to know if one is more effective then others.
I use A1111 for controlnet openpose, adetailer and reactor, does Forge have those tools? I would like to use the Flux model.
They reallyyyy need animatediff working in Forge with all the speed improvements in place
animatediif works on forge since the beginning, of the resistance....
As stated above, Animatediff does work. You may need a specific fork of it but it's out there.
Yes Forge is lacking in video extensions.
Invoke got a nice update lately tho. And it's like on the another end of the UI/UX principal to the ComfyUI.
It's solid and easy to use. Artist friendly. Also offer some level of node workflow which isn't as powerful as Comfy but instead make someone that knows the nodes could make an automated tool out of it for a non-node user. Which the interface also offer to hide the node and see only the selected inputs and the user just press the generate button to run that workflow.
Does the local/community version have feature parity with the priced options? I remember seeing Invoke a year or so ago and was really impressed with its ux/ui. The AI scene is just f--king exploding constantly and I want just one tool I can really settle into and learn really well. I know ComfyUI is probably the most robust toolset but its ux/ui looks like a nightmare. At that point I'd rather just paint the actual images mysef.
Invoke is God tier Inpainting outpainting. Wish Forge could combine with Invoke for that alone.
I've been using forge with flux for a couple of weeks now. It's perfect for what I need it for.
@@bemusedkidney8619 does it support "Control net" and "LORA" ?
Does it support "Control net" and "LORA"?
Lora is working perfectly. You just need to activate fp16 lora from the top tabs
@@kiransurwade3576Lora yes, controlnet no, not yet.. unless I've set it up wrong lol
So i download loras for flux on civitai but idk if theyre working? In forge.
hi, where can i enable: vae/ text encoder? i don't have it :(
Great video! I'm a long time A1111 user, and I just installed forge 2 weeks ago. It's really, really good, but It's not perfect. It seems to have some UI issues, you can't use the refiner currently, and it maybe has some issues handling memory on occasion. But it can use Flux models with ease, and it seems to render and upscale faster than A1111 for me at least. I'm keeping both.
Very nice… your recommendation is meaningful. Can it do LoRA training?
Thank you! I just did this, but didn't select any vae or text encoders and it generated fine. Maybe those are needed only for normal dev? Seems like they're integrated in the nf4 v2?
You're right.
I installed Forge a week or so back. I never have any luck installing as snag free as youtubers do! I do recall I had to upgrade my CUDA which was straightforward but be careful to upgrade to the level of the forge install you are using. Its a good idea to carefully watch the execution window for clues as to whats gone wrong (often models don't download completely and you have to manually do the download again) If you keep getting errors restart forge and then copy the whole dialog from the window into perplexity or gemini to get advice.
Its not a waste of time to still be using A1111.
After trying both Im still using A1111, forge might have been slightly faster for each image (less than 10 seconds) but it comes with a bunch of extensions i don't use and doesn't support others that i do.
Also A1111 hasn't broken on me yet wile my forge stopped opening out of no where so im not sure whats up with that
I've used Forge for quite a few months. Good to see it in the limelight.
What about SwarmUI ? I find it even more stable and organised, can you try it please and cover it in a video ?
Switched to Forge recently. Works nice, just a bit disappointed with the Flux model. And I'm still not able to activate Clip Skip for XL/PONY. Or is it this Clip Skip for SDXL checkbox? Anyway. If you don't want to play around with Flux, you can still use A1111.
I've been using for months.... there are some downsides with some plugins, but usually works pretty well. It renders SDXL LCM models in less than 10 seconds on my 4GB VRAM
@@eduMachado83 what gou though? I can gen sdxl on my 980ti 6GB but it’s more like close to a minute (which is wild because originally it took like 6 minutes in comfy)
Forge has moved to beta forever and we be a testing ground for features that's going to be ported to A1111 after testing. So it will always be ahead of A1111 in terms of commits but A1111 will always be more stable
A1111 nothing has happened for the 2nd month, there is no information about the next release, the project seems to be frozen
@@havemoney ya I was gonna say, isn't A1111 a dead project?
@@Exaltar A1111 and Forge are convenient for me, what will happen with A1111 will we are waiting for information from the developer.
Just a heads up for people having extension issues in forge, if forge gets stuck at 50% when generating its because your dimensions are not divisible by both 8 and 32.
In forge this is common because the slider goes in 8 by 8 increments.
All numbers divisible by 32 are divisible by 8 but not every number divisible by 8 is divisible by 32, that's the problem.
If you want to avoid the hassle of calculating, under the config file change it so the increments of the dimensions sliders are 32 by 32 and you won't have that problem just be careful with the upscale make sure the result is also divisible by 32.
This is useful specially if you use regional prompter and other extensions, they won't crash anymore
For my needs most extensions work, except for composable lora advanced functions but i honestly don't need it.
I think I'll learn Comfui just for Flux. Currently the generation speed on Flux for upscaling is significantly better compared to Forge. Upscaling is essential for me.
Thank you, Olivio👍
am getting the error message: "connection errored out" when I try to generate a picture or do anything. Anyone know how to solve this? I've got an rtx 4090 and 32gb ram, it should be enough no?
I'm getting You do not have CLIP state dict error. what should I do?
Is it necessary to install these VAE? Will the result be better with them?
Hi, the only thing that interests me, is not covered by your very informative video: DEFORUM ?????
@@M--S I got deforum to (finally) install on the latest Forge, but it was very very painful
@@MrPlasmo I've already heard that before. Do XL-controlnets work? Do you have any link to a tutorial how to install it?
I just switch to Forge after spending almost 2 years on A1111 , took almost no time to set it back
My laptop with 4gb GPU couldnt handle SD.XL on A1111 but it can on Forge , its a little bit slower than SD1.5 but it give me way better results.
i definitly dont regret installing Forge
thanks for the video , i install a second forge for flux
Automatic1111 supporter through every error message and crash
Me too, still best for X/Y script Prompt S/R
I tried Forge last time Olivio released a video about it, but I didn't see any speed improvement over Auto1111. At least no visible one.
its never crashed for me, maybe you could say I dont use many advanced features often.
@@SupremacyGamesYT yeah, forge hasn't crashed for me either.
Ok. I go with Stability Matrix. It starts to be a little mess here :) I'll try it definitely for low VRAM Flux.
Forge still doesn't support flux lora. When i use flux models, forge keeps moving the model between my RAM and vram, before every generation, despite that i have 16gb vram. Which makes it super slow.
It supports lora. Include nf4. Just need to activate the fp16 lora from the top menu. Then add your lora. Annd yes its slow then comfy. 38seconds vs 1 minute with my 4060..
Err, yes it does support Flux lora. As well as fp16 stated above, you may also need to drop the GPU weights value down (in the 8000's is a figure I've seen suggested). It isn't quick but it definitely works because I'm using them with 12Gb VRAM. Slow doesn't mean not working.
@@UgurDoydukFM bummer, i get FASTER RESULTS with FORGE on my 6GB rtx 2060 using the regular fp8 DEV model. At the suggested 896x1152 i can do 20 steps in 1:30 seconds! and a 1080x1440 image in 2:30 mionutes!! how come you have a BETTER gpu and almost the same as me?
Loras do work for me. though. And I only have 12 Gb
Forge is now a funnel to test features for A1111. I don't know why you don't mention this. A1111 is the way to go if you don't want to be a tester of new features that might/might not work.
This video is a good reminder as to why I left the toxic and misleading community Olivio runs.
The way you speak reminds me of how "Christopher Waltz" speaks in Django Unchained and Inglorious Bastards!
You know you could script up and distribute docker compose files with a similar amount of effort as editing one of your batch files? After a one-time setup of WSL2, Docker Desktop (free), and NVidia's Container Toolkit, people could run your script to download everything into a ready-to-go container (like a lightweight VM) that will allow them to run the stuff while also insulating them from the possibility of malicious software. Software like this is a perfect fit for containers.
Sounds cool, but i have literally no idea what you just said. So it's cool that that works, but you need to know how. And at this moment in time i don't
yeah I've switched completely to forge and swarmui for my daily gens.
A1111 memory management sucks, very very bad.
for 1.5 it's not that bad, but for pdxl and sdxl yeah it's bad
Are AMD gpus supported?
Hi, to generate video (img to video) how do you do it? Which model? Which extension?
I was thinking of moving to forge UI from fooocus for more options, should I do that and how much VRM would I need to run flux
@@pieugedher257 I did exactly that several months ago and never went back. I have 6Gb vram and running flux with no problems.
I'm using forge for some days and I'm liking it.
For me it's too rigid. While comfyui is far from perfect, it's a lot more flexible and allows you to pick up quickly where you left off various projects.
It depends on what u need. I use both
@@AgustinCaniglia1992 I also use both although I definitely prefer Forge for most things.
I replaced a1111 and forge with Krita ai extension for convenience and control.
It runs on comfyUi but it's much less complicated for image control and tweak.
i just installed it and i can't manage to get an image without getting an error ! it seems that forge is not ready enough to be sent on the podium !
where were you all this time? Forge is amazing (again) since the first days of Flux release
I think ComfyUI has a really steep learning curve to start out with, but once you understand it you're good. Good to know there's an actively developed successor to A1111 though.
Thank you! ❤
I don't know why it didn't get mentioned or I missed it - Forge is mostly a downstream version of A1111 webui ("Forge is a platform on top of Stable Diffusion WebUI") with stuff added to it, so it wouldn't probably wouldn't even exist without the work put into the original webui. Also, there is no point of moving or copying large model files, better create one big folder where you store the models, then create symbolic links into A1111 and Forge folders. That works nicely when you have 5+ different UIs for SD, so you don't have to move models around and you don't have to bloat your installs.
Should I use Comfyui? A1111? Forge? I think I watched the exact same video but instead of forge it's comfyui. Which not gonna lie, I still like comfyui way more than A1111. But is there a better UI already than comfyui? Honestly, the only complaint I have with comfyui is that it's quite unstable. But it does a really good job of experimenting with new ways of generating ai art.
I am still using ComfyUI , where do I stand on the dead horse betting spectrum?
right now I am using Fooocus for stable diffusion. how to link my Fooocus checkpoints folder to Forge UI?
noob here. do i need to worry about git when having both a1111 and forge on my machine? will i get any git directory errors? would it be wise to uninstal any and all software related to a1111 before changing over to forge'?
For some reason, mine will only show the VAE and not the VAE and the Text Encoder
you are very wrong, some crucial extensions are broken in forge , this makes it dealbreaker
Is it the same of What we get via pinokio 2 or it's different / updated version ?
Dont use pinokio. Its not good.
Is it normal that it takes forever to extract files after downloading One Click package? Iv been sitting for over 40 min and files er still extracting...
why some other sampling method not present on the forge ui like DPM++ 2M SDE Exponential
Why not just do it via stability matrix, you've covered this? Most of the config file editing is done for you. And no need to move models.
Using it atm because of Flux, BUT there are clear downsides: It can't do proper hashes for cross-posting on Civitai (Have to manually replace the garbage hashes with the proper ones in each image and I frickin hate it). Several extensions, that work in A1111 don't work in Forge, so yeah ... cool it a little with the hype, maybe?
Hype? Forge has been out for months. It's faster than a1111 and made by the same person
@@n3tw0rk_n3k0 Do you know the definition of hype? Doesn't look like it
Good video I like your videos. Unfortanly, I had some less positive experiences after that; they make it flux-compatible I have regular black images, while the version below it work very stable, be a ware I am running on mac m1
Hi ! Hello ! they is no SVD ? where is the svd tab ?? thank you !
The king will always be, Lord ComfyUi
YES!!!!!! Finally some common sense!!!!
good luck filling hardrive with tons of png files.
Next, please make a video on Forge UI vs Comfy UI.
Hello, I have 35 folders in my models in ComfyUI and there is only 6 paths in your script, is it normal ? A lot are missing like unet, ipadapter, inpaint, etc…
At least in regards Colab Notebook versions of Forge--there is still no support for Controlnet batch processing. Forge only processes the first image. Been an unresoved bug going on seven months.
It takes me 21Minutes for one image with my 6gb VRAM! Any idea why?
set COMMANDLINE_ARGS= --no-half --lowvram --opt-split-attention --xformers --api
currently using this IN A1111 and do i need to use lowvram on forgeUI? 4gb VRAM user here
Forge UI was abandoned, and then restarted with flat out comments by the author that he was building it for himself and not to expect it to be useful for production use. For that reason alone I'm very hesitant to use it. A1111 has its issues, but it hasn't been abandoned. I do use it for Flux, but that isn't my primary model right now.
Hi, it is Mehdy. I would like to know if you can test Krita with Flux ans its Controlnets. Thanjks.
Indeed, Krita is m'y go to UI ❤
Putting SD_VAE in my quicksettings in forge ui didn't actually put the SD_VAE in my quick settings...
Can you use xformers with this?
No for webui_forge_cu124_torch24. Not sure about webui_forge_cu121_torch231
you lost me at the *quick list* could not find it ? il figgure it out though..
Is there a linux implementation of this, or can I branch from A1111 to forge UI?
I switched from A1111 WebUI to Forge WebUI because Forge performance is faster than A1111 with RTX 3050 8GB GPU when running SD1.5, SDXL, and PDXL models. I ran the Flux 1 dev model with Lora takes 1-3 minutes for 1-2 images. If without Lora it can take 30-60 seconds
PDXL models?
i'll stick to comfy ♥♥
I find the spaghetti monster too time consuming if you want to do anything even remotely complicated.
@@jimg9820Krita ai for simple use or specific artistic control. comfyUi for automated/precice problem solving.
In both case it's comfyUi as engine.
@@jimg9820 While it may be harder to pick up initially, when you do pick it up it is genuinely unfathomably more convenient and you’re able to do so much more than a gradio ui, that came out what, 4 years ago now?
Can't find how to choose the wieght for my loras in forge
I tried forge for FLUX but it seems unstable on my computer, while SwarmUI works fine. The only issue with SwarmUI is that it doesnt save the image metadata in the images themselves.
I modified the webui-user.bat exactly like in the video but its not linking with automatic 111 models neither opening ForgeUI...
I had to change my slashes for the directory from backslash \ to forward slash / I've read in some cases it needs to be double [//] and your a1111 install location may be different so do not copy his exactly.
@@Jakeurb8ty82 I put the double[//], now its opening ForgeUI, but models are not available, I put the right address, in my case, C://pinokio//api//automatic1111.git
i loaded it and Deforum is NOT showing up as a tab
hello when i launch forge does it always update or do i have to launch the update.bat once in a while please ?
is there a way to use this with my m2 macbook? thx in advance!
Tried this and then uninstalled it not long after, I like to make img2img animations and guess what? img2img batching is completely broken on forge, all it does is just generate the first img of a batch over and over and over again, and the developer doesnt look like they have any plans to fix it anytime soon. To save you all the trouble, stick to regular a1111.
the train tab is mysteriously missing, but training is still a sub menu in the settings
It's not a hate speech BTW, but Forge 2 is a personal project never be done, go inspire the community to help this guy!
I'm happy this is not more comfyui... I tried but it makes me unhappy
A1111 is still all I'm confident with but I'll give this a try
I installed Forge via Stability Matrix. But in some Models there is a "Clip" error. I use rtx 3070 and usualy it works fine.
winrar isn't free, 7z is free, winrar is shareware :)
what means shareware?
Yeah, I have no idea why people think 7zip isn't free. It's always been free and unlike winrar, it isn't like 30 years old shareware.
@@FutonGama It means you are supposed to pay for it, it's not free.
@@arcanescroll gotcha, i use both but i'm used more to 7zip, also i have a Dark Theme for it
Peazip is better
Isn't Forge just an updated Automatic 1111???
With a1111 rocm works with comfyui and forge not much :(
Good day sir! I'm very new to all this SD thing. One issue I encounter is that when I create images with my favorite model, it only works well at a specific resolution. When I increase the resolution, I face the problem of duplicated characters. Is there any way to solve this issue? How can I create an image with my favorite model at the desired resolution? Thank you for taking the time to respond!
Not the OP, but check if the UI has hires fix. Generate the images at the resolution they're most ideally generated from, but turn on hires fix and change the scale to upscale after the first generation.