I should also note that we have an extensive list of environment variables that you can set to automatically take care of things like your model paths, output paths, debug logging, etc. There's dozens of them, all on the wiki with our cli arguments. So for example if you set `SD_MODELSDIR=D:\Models` in your environment, even a fresh install would immediately look there on startup for your existing models of all kinds.
@@cleverestx thanks for the reply. I took the time to watch Sebastian's video but I'm still not sure what makes it better than A1111, it looks like it does the same thing with (almost) the same UI.
There's a whole list of things it does out of the box that you need 30+ extensions installed for in A1111, our parser works properly, and it's just plain faster if you put in a little effort to tuning it for your needs.
Use A1111 if you have high end gpu, like 12GB vram and higher. Use Forge Webui, if you have lower end GPU, like 6GB to 8GB vram. I use defooocus (a fooocus fork) specifically for inpainting and outpainting, as it was doing the best job at that specific task. Use ConfyUI if you want a more advanced and most flexible ui.
Thanks for the video, but, why woukd i want to install this ui in the first place? I mean it's still sdxl and 1.5, so what is so great about this sdnext? That forge or comfyui don't have? Too many ui's, why shoukd i care? Asking seriously
Difficult to say, that's going to depend on your hardware capabilities, but we do not enable a lot of performance or memory management options automatically, they're in the settings so you can adjust them as needed for your situation or task.
The errors Sebastian encountered regarding the packages were reporting that the versions were wrong due to package conflict issues, the correct version was installed right after the warning. Those are informative, not actually an issue a user needs to worry about.
Hi Sebastian. I am running a Mac M1 Max. wondering which stable diffusion ui I should download that will run well locally? Comfy UI, SDNext, Forge, Control Next? Please lemme know if possible.
So can anyone help me out with this one, i did the guide, it all looks fine and dandy but im not sure im even laoding models, my generate button says "txting2img starting" And im not finding an answer to this one
I tried using this. It was very slow. Gave me errors constantly. For some reason, when I closed the preview window for my models, there doesn't seem a way to bring it back? I use a1111 and Swarm a bunch. Never had this many problems (swarm is crazy fast). Maybe it's just me, but this seems like it needs a LOT of work.
Hallo Sebastian, thanks for the great tutorials. Could you please do one on using ControlNet for Architectural rendering exteriors and interiors from sketches? thanks alot!
Hi Somar! I have multiple controlnet videos that you can watch. Even though the examples might be humans, the process will be similar for architecture.
That was a nice video, thanks. I might use SDNext, it looks reasonable, but actually the UI doesn't matter much to me. I could get along with most UIs, if only the models would be doing what I want! If I am very lucky, then SD3 will come for the rescue, but I doubt it. In the end, I want to create something novel, and all models which I have seen can only reproduce their training sets, with some (small) variations. Since my ideas are not in the training set, trying to create images with Stable Diffusion means it will forever ignore my prompts and just reproduce something from its training set instead.
First time trying out a locally running image generator. Managed the generate one image, and not working since that. :( No progress bar, no error message, nothing...
With this new update they've addressed a lot, so it's much better than the previous versions. Still, not my first choice. I do love that they're trying to improve on the "standard" SD UI though.
@@jgt76 overall feel of that webui as a tool is not better than Forge - I'm not talking about the UI, but about the way it is working with SDXL image generation - not impressed at all
Sdnext is fairly buggy. Can't get an img2img to upscale pass x1 scale. By the way you can add labels to the icons in the modern theme so you don't have to hover over them.
Thank you for this video. I was not able to use SDnext without this explanation. I was trying to install AUTOMATIC1111 but it has a bug now and does not work with my graphic card, also comfyUI does not work with my PC configuration now, so SDNext must have something better in installation.
It is a custom installation that handles things dynamically as best it can, if you select things in the UI, it will install or uninstall libraries for them, such as xformers. As well as automatically downloading models you need along the way to do things like face swapping, ip-adapters, and a full suite of controlnet models.
It's portable as far as I know, if you were to set an environment variable for the location of your python install, it would always use that, so you could presumably put it on a usb stick or something.
Sir, please, please i request 🙏 make a videos "rope pearl" latest version & installing process, i have try but a lot of error,😢 (anaconda prompt, python 3.10 , VS studio) how to set, steps, please guide Sir
Thanks for the tutorial. I tried a few times, but I don't know what the advantage is to AUTOMATIC or Forge...
I should also note that we have an extensive list of environment variables that you can set to automatically take care of things like your model paths, output paths, debug logging, etc. There's dozens of them, all on the wiki with our cli arguments. So for example if you set `SD_MODELSDIR=D:\Models` in your environment, even a fresh install would immediately look there on startup for your existing models of all kinds.
I have no idea what SDNext is
A better version of Automatic1111, used to be a fork, but not exceeded it in seemless model support and performance.
@@cleverestx thanks for the reply. I took the time to watch Sebastian's video but I'm still not sure what makes it better than A1111, it looks like it does the same thing with (almost) the same UI.
Thanks for the tutorial. I would wish for a comparison, like I haven't seen anything that I couldn't do with A1111.
There's a whole list of things it does out of the box that you need 30+ extensions installed for in A1111, our parser works properly, and it's just plain faster if you put in a little effort to tuning it for your needs.
Which interface do you recommend the most? Which one do you think is better? Automatic1111, Forge, Invoke, Vlad Automatic, Foocus?
The ones I use the most are: a1111, Comfy, RuinedFooocus. I want to get into Stable Swarm more too.
Why a1111 when Forge is faster with same ui?
Use A1111 if you have high end gpu, like 12GB vram and higher.
Use Forge Webui, if you have lower end GPU, like 6GB to 8GB vram.
I use defooocus (a fooocus fork) specifically for inpainting and outpainting, as it was doing the best job at that specific task.
Use ConfyUI if you want a more advanced and most flexible ui.
I'm using Forge as a main interface. Automatic1111, Foocus as secondary. Comfyui is for experimental functions.
@@sebastiankamph What is good about RuinedFooocus? I found it way worse than ordinary Fooocus.
Thanks for the video, but, why woukd i want to install this ui in the first place? I mean it's still sdxl and 1.5, so what is so great about this sdnext? That forge or comfyui don't have? Too many ui's, why shoukd i care? Asking seriously
Better performance. I guess it's for those who want something like A1111, but without the whole "keep your VRAM hostage" bit.
Can you do a tutorial on the control net feature of sd next? I find it pretty confusing.
how is it performance wise compared to forge?
same question
same question
same
Difficult to say, that's going to depend on your hardware capabilities, but we do not enable a lot of performance or memory management options automatically, they're in the settings so you can adjust them as needed for your situation or task.
Shame question
The errors Sebastian encountered regarding the packages were reporting that the versions were wrong due to package conflict issues, the correct version was installed right after the warning. Those are informative, not actually an issue a user needs to worry about.
Any chance you might cover its equivalent to controlnet? I swear I get lost trying to navigate that.
Hi Sebastian. I am running a Mac M1 Max. wondering which stable diffusion ui I should download that will run well locally? Comfy UI, SDNext, Forge, Control Next? Please lemme know if possible.
Why it denies access when I put the githut directory on cmd?
So can anyone help me out with this one, i did the guide, it all looks fine and dandy but im not sure im even laoding models, my generate button says "txting2img starting"
And im not finding an answer to this one
same here
I tried using this. It was very slow. Gave me errors constantly. For some reason, when I closed the preview window for my models, there doesn't seem a way to bring it back?
I use a1111 and Swarm a bunch. Never had this many problems (swarm is crazy fast). Maybe it's just me, but this seems like it needs a LOT of work.
Hallo Sebastian, thanks for the great tutorials. Could you please do one on using ControlNet for Architectural rendering exteriors and interiors from sketches? thanks alot!
Hi Somar! I have multiple controlnet videos that you can watch. Even though the examples might be humans, the process will be similar for architecture.
That was a nice video, thanks. I might use SDNext, it looks reasonable, but actually the UI doesn't matter much to me. I could get along with most UIs, if only the models would be doing what I want! If I am very lucky, then SD3 will come for the rescue, but I doubt it. In the end, I want to create something novel, and all models which I have seen can only reproduce their training sets, with some (small) variations. Since my ideas are not in the training set, trying to create images with Stable Diffusion means it will forever ignore my prompts and just reproduce something from its training set instead.
We do have two UIs, one is gradio, more like A1111, and this new modern ui.
First time trying out a locally running image generator. Managed the generate one image, and not working since that. :( No progress bar, no error message, nothing...
If you aren't seeing anything on your console, then are you sure you're running it at all?
I'm wondering why is it so slow?
It takes an hour for the opening, while a1111 - a few minutes.
Does somebody have the same problem?
That's not normal, if anything we start up faster, our start time can get down to the 10-15s range with a couple of settings changes.
There is new controlnet named "Starline" to colorize lineart. Its not integrated with WebUI yet but I hope to understand how it works
Is it up on huggingface? I'll have it integrated if it hasn't already been, I think we did just add a new controlnet or two the other day.
sdnext was "promising" but after testing it ---- nah
Snap
With this new update they've addressed a lot, so it's much better than the previous versions. Still, not my first choice. I do love that they're trying to improve on the "standard" SD UI though.
I'd win
If you have specific feedback we're always happy to hear it and we respond to it quickly as possible. This new ui is a preview, it's not done.
@@jgt76 overall feel of that webui as a tool is not better than Forge - I'm not talking about the UI, but about the way it is working with SDXL image generation - not impressed at all
Sdnext is fairly buggy. Can't get an img2img to upscale pass x1 scale. By the way you can add labels to the icons in the modern theme so you don't have to hover over them.
That's an excellent tip!
Yes, it is amazing what one can find if they look in the Settings! 😆
Thank you for this video. I was not able to use SDnext without this explanation. I was trying to install AUTOMATIC1111 but it has a bug now and does not work with my graphic card, also comfyUI does not work with my PC configuration now, so SDNext must have something better in installation.
It is a custom installation that handles things dynamically as best it can, if you select things in the UI, it will install or uninstall libraries for them, such as xformers. As well as automatically downloading models you need along the way to do things like face swapping, ip-adapters, and a full suite of controlnet models.
You know that stabilitymatrix exists, right?
Is there a portable version?
It's portable as far as I know, if you were to set an environment variable for the location of your python install, it would always use that, so you could presumably put it on a usb stick or something.
i wish this all is for A1111 not crap useless comfyui or any other..............................
Sir, please, please i request 🙏 make a videos "rope pearl" latest version & installing process,
i have try but a lot of error,😢
(anaconda prompt, python 3.10 , VS studio) how to set, steps, please guide Sir
Can you please give your mouse icon theme
i cant find any models or files in final step models\Stable-diffusion location? can you tell me any solution?
How does this differ from regular stable diffusion?
Is you controlnet working?
Is it worth installing this instead of using the normal Stable Diffusion? Any major difference between the two?
Matter of preference. If you like a1111 or Forge but want a new UI for it, this could be interesting for you.
@@sebastiankamph that is one of my biggest gripes for normal so might give it a try thanks!
👋