bye midjourney! SDXL 1.0 - How to install Stable Diffusion XL 1.0 (Automatic1111 & ComfyUI Tutorial)

Поделиться
HTML-код
  • Опубликовано: 29 янв 2025

Комментарии • 130

  • @CoderXAI
    @CoderXAI  Год назад +26

    thanks for watching! might be live on twitch for debugging, questions and chat: www.twitch.tv/coderx
    huggingface sdxl1.0 base model: huggingface.co/stabilityai/stable-diffusion-xl-base-1.0
    huggingface sdxl1.0 refiner model: huggingface.co/stabilityai/stable-diffusion-xl-refiner-1.0
    automatic1111: github.com/AUTOMATIC1111/stable-diffusion-webui
    comfyui: github.com/comfyanonymous/ComfyUI
    refiner.json (now updated to refiner_v1.0.json) by camenduru: github.com/camenduru/sdxl-colab/blob/main/refiner_v1.0.json

    • @dynoko3295
      @dynoko3295 Год назад

      you should probaby pin this

    • @CoderXAI
      @CoderXAI  Год назад +1

      @@dynoko3295 I thought this was always pinned, this explains a lot of comments :(

    • @DT-GDS
      @DT-GDS Год назад

      Looks like they took down the SDXL model...

    • @CoderXAI
      @CoderXAI  Год назад

      looks like they updated the base+refiner models, there were some issues with VAE so they are probably(hopefully) better now

    • @DT-GDS
      @DT-GDS Год назад +1

      @@CoderXAI I do not see the tensor model in that link. Is it somewhere else?

  • @globalmodelsdotbiz8965
    @globalmodelsdotbiz8965 Год назад +7

    Clipdrop error: Stable Diffusion XL (Watermark, 400 images per day) not per month for free users.

  • @haroldrandolf2981
    @haroldrandolf2981 Год назад +2

    Hey CoderX,
    I have been trying to generate some ummm spicy images but I cant seem to. Im using ComfyUI coz i cant run Automatic1111 or Vladmandic.
    Is it ComfyUI's problem?
    Also I used absolute reality and its generating what i want but it is again censoring my images if i try to send it to refiner.
    Can you please help me?
    Thanking you,
    Yours faithfully,
    Harold

  • @creedolala6918
    @creedolala6918 Год назад +2

    Appreciate a guide that is not over-explained or under-explained. Was curious about comfy after finding that it seems like it avoids out-of-memory errors, while a1111 crashes with this model. I guess I shoulda got more vram.

  • @THISISMYHANDLETODAY
    @THISISMYHANDLETODAY Год назад +8

    Just came from another video trying to sell a one click download LOL. Man, thanks for the quick and concise tutorial!!!

    • @cl4911
      @cl4911 8 месяцев назад

      yep SEcourse's video was suggested right after this lmaoo

  • @scottgust9709
    @scottgust9709 Год назад +4

    Where do you put the Refiner Model in Auto1111?

    • @tal7atal7a66
      @tal7atal7a66 Год назад

      same place " path " for the basic model .

  • @effinballers2543
    @effinballers2543 Год назад +2

    i have 6gb vram, but there is an error while running update bat. How can i fix it? it says:
    "return t.to(device, dtype if t.is_floating_point() or t.is_complex() else None, non_blocking)
    torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 20.00 MiB (GPU 0; 6.00 GiB total capacity; 4.06 GiB already allocated; 14.71 MiB free; 4.14 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF
    Stable diffusion model failed to load
    Applying attention optimization: Doggettx... done."

    • @larrog8413
      @larrog8413 Год назад +1

      I have 8gb and same problem

  • @kuvjason7236
    @kuvjason7236 6 месяцев назад

    Help When I make an image .. i get an error with *Folder "outputs\txt2img-images" does not exist. after you save an image, the folder will be created.* idk how to fix this

  • @Sergatx
    @Sergatx Год назад +4

    Where did you get the refiner.json from?

  • @ParvathyKapoor
    @ParvathyKapoor Год назад

    Does deforum support?

  • @s.a.1083
    @s.a.1083 3 месяца назад

    I see outputs are saved in the webui folder. Is there any chance that the prompts used to create them are saved somewhere too?

  • @Hugh_Mungus
    @Hugh_Mungus Год назад

    XL models won't load on my A1111 UI, it's not the gpu and I've tried reinstlaling, updating, etc

  • @Oismyurl
    @Oismyurl Год назад

    I can not get the refiner to work... I keep getting ERR reconnecting... With both 0.9 and 1.0. I have tried to update but that didn't fix the issue.

  • @yasen_stoev
    @yasen_stoev Год назад +5

    anyone getting the error: RuntimeError: Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check

    • @CoderXAI
      @CoderXAI  Год назад +2

      could be some sort of wrong torch error bug, here's a relevant link: github.com/AUTOMATIC1111/stable-diffusion-webui/issues/9402

    • @TheBlackFisk
      @TheBlackFisk Год назад

      Been getting that on 1.4 update and even after a clean install of 1.5

    • @RoboMagician
      @RoboMagician Год назад

      i run into this problem when using refiner

  • @chazanderson4532
    @chazanderson4532 Год назад +4

    Any advice for getting this working with AMD GPU?

    • @Dragon211
      @Dragon211 Год назад

      I’d like to know as well

  • @JSwanson547
    @JSwanson547 Год назад +2

    When using the refiner, do both models occupy VRAM similtaneously, or does the base unload to offer more space to the refiner?

    • @HolidayAtHome
      @HolidayAtHome Год назад

      Base unloads and then the refiner loads in ;)

  • @aurum.graphics
    @aurum.graphics Год назад

    I have this error: Stable diffusion model failed to load
    Loading weights [31e35c80fc] from C:\Users\Documents\Stable Diffusion\Webui2\webui\models\Stable-diffusion\sd_xl_base_1.0.safetensors

  • @christoff124
    @christoff124 Год назад +1

    what folder does the refiner go into?

  • @wasted828
    @wasted828 Год назад +1

    where do you put the refiner file in the 1111 webui folder?

  • @EuphoricDreamSequence
    @EuphoricDreamSequence Год назад +5

    Excited about 1.0. Going to try today.
    Btw the background music is wonderful
    And your explanation is clean, clear and to the point.

    • @CoderXAI
      @CoderXAI  Год назад +1

      thank you, you're too kind! I had been planning this video for over 2 weeks now since SDXL 1.0 was supposed to launch on 18 so it feels good that it has been helpful to others :D

    • @DanVogt
      @DanVogt Год назад

      @@CoderXAI I agree, you have done a great job making this accessible and easy to understand. Thank you so much. May I know what the music is please?

    • @FirstLast-tx3yj
      @FirstLast-tx3yj Год назад

      ​@@CoderXAIhello can I batch modify 10 frames like we img2img like we use to do in the old stable diffusion?

  • @wrillywonka1320
    @wrillywonka1320 Год назад

    i keep getting this message "Creating model from config: C:\Users\Admin\stable-diffusion-webui
    epositories\generative-models\configs\inference\sd_xl_base.yaml
    Failed to create model quickly; will retry using slow method." do you know why?

  • @aurum.graphics
    @aurum.graphics Год назад

    when installing run.bat i got out of space, i had to remove folders to make space in the memory. Now i finnished installation but when I launch SD it says "error" when i press generate button, :( solutions??

  • @Im_that_guy_man
    @Im_that_guy_man Год назад +1

    is it imposibble to use the refiner in automatic 1111?

  • @Nuverotic
    @Nuverotic 3 месяца назад +1

    I click generate and nothing happens

  • @Bigblackninja5
    @Bigblackninja5 Год назад

    How do I do the options that were in automatic111 inside comfyui like inpainting, imgtoimg ect

  • @aiwithoutsecrets
    @aiwithoutsecrets Год назад +2

    Was able to install and run with both interfaces, thank you

  • @kevinehsani3358
    @kevinehsani3358 Год назад

    I am trying to use automatic1111 and sdxl-refiner-1.0 and have memory issue, is there a way to set it up to use cpu since most of my gpu memory is reserved by pytorch. This is the error I get, it loads up but can not run a prompt "Tried to allocate 64.00 MiB (GPU 0; 8.00 GiB total capacity; 7.20 GiB already allocated; 0 bytes free; 7.33 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation."

    • @hernanpulgar7237
      @hernanpulgar7237 Год назад

      I use A1111 with CPU only, and prompts that take 4-5 minutes with v1.5, now run 5 hours with SDXL (and just 512x512, have not tried higher resolutions yet!)
      You can increase the memory as much as you want (physical is better but swap/virtual should be ok), but memory *is not* your real problem if you use only CPU for SDXL.

  • @rafa-lk6lf
    @rafa-lk6lf 10 месяцев назад

    don't know why\how but my comfyUI works way slower than auto1111, does comfy need more vram to generate images or something like that?

  • @DARKNESSMANZ
    @DARKNESSMANZ Год назад

    im using vlad diffusion the sd xl model was Laoding then stopped loading at 70 percent

  • @bladechild2449
    @bladechild2449 Год назад +1

    I don't really get why comfyui seems to generate images, with refiner, in like a minute on an 8gb card, but it takes like 6 minutes in A1111

    • @Nawilżone
      @Nawilżone Год назад

      i wont be using this it takes on my card to generate one image 512x512 like 5 minutes what the hell i can make like 12 images like that on Rev animated in 1minute

  • @legioneelletregi1100
    @legioneelletregi1100 Год назад

    when I try to load the refiner json nothing happens. I downloaded the updated version btw

  • @foxyfox7627
    @foxyfox7627 Год назад +1

    Can you install in on the existing sd?

  • @ippibean
    @ippibean Год назад

    Was your image gen sped up? I have a 3060 12gb vram and it takes about a minute for me with base+refiner.

    • @CoderXAI
      @CoderXAI  Год назад +1

      same graphics card, around similar speed (~40-50s)

  • @cy7
    @cy7 Год назад +1

    What do you think about Happy Diffusion

    • @sirtimatbob
      @sirtimatbob Год назад

      What do YOU think about Happy Diffusion?

  • @iloveshibainu9003
    @iloveshibainu9003 Год назад +1

    can we use all automatic 1111 ckpts in in comfy UI ??

  • @mistertitanic33
    @mistertitanic33 Год назад

    I have 2070 RTX and 16GB of ram but I keep getting OutOfMemoryError: CUDA out of memory. I have xformers installed and I turned Token Merging ratio up but I still get the error. Any idea how I can resolve this. Using Automatic1111

    • @AioWey
      @AioWey Год назад

      2080 super here, same

    • @ryand-rs7pu
      @ryand-rs7pu Год назад

      i have 2070 too, xl and a111 no work ,but comfy work fine

  • @ZenBenzineX
    @ZenBenzineX Год назад

    hi, when I upload the model, give a prompt and generate, the webui doesnt even move. Am I doing something wrong?

    • @CoderXAI
      @CoderXAI  Год назад

      does it say anything on the terminal? it'll either throw an error or show what it's loading/doing

    • @ZenBenzineX
      @ZenBenzineX Год назад

      @@CoderXAI That is the thing, no error message. it just stays with just the text that come just before the image starts rendering. and stays there forever.

  • @TeamLiftMedia
    @TeamLiftMedia Год назад +4

    Great explanation and guide. Thank you.

  • @palomaetienne
    @palomaetienne Год назад

    What's the difference between the refiner and the base model, please?

    • @CoderXAI
      @CoderXAI  Год назад +1

      base model is what generates the image/the main model
      refiner is an additional model that takes the generated image and adds more details to it(so kind of optional)

    • @palomaetienne
      @palomaetienne Год назад

      @@CoderXAI Thank you for your explanation 🐬

  • @danielhernanalonso7219
    @danielhernanalonso7219 Год назад

    Noob question. Why do you make a new installation of automatic1111 from a previous build instead of simply adding the SDXL model to the automatic1111 that you were already using?

    • @CoderXAI
      @CoderXAI  Год назад +1

      if you already have existing automatic1111 and you can update that using update.bat script or manually, you don't need to do a fresh install. you do need to update since A1111 was recently updated to support SDXL and older versions won't work with it

  • @xcrashqq
    @xcrashqq Год назад

    I wish there would be full explanation how to install it from very beginning, with this other app that you had before.

  • @RomboDawg
    @RomboDawg Год назад +1

    (EDIT: I can confirm this issue is fixed with the new update of automatic 1111)
    I followed your previous tutorial and everytime i launched automatic 1111 it would redownload all the pytorch.bin files ever single time and take like 10 minutes to launch the web ui. I really hope that doesnt happen this time, but is there a way to prevent this from happening?

    • @RomboDawg
      @RomboDawg Год назад

      Im still downloading the files so this issue might be fixed, but im still yet to see. ill update you

  • @MrHannessie
    @MrHannessie Год назад +2

    Pfff, Bye Midjourney? I dont think so. Everyone saying MJ killer. Its not. Both have pros and Cons. MJ still usually looks better. SD XL is an improvement sure and you have more control and can do nsfw, but it still cant compare to how many MJ images look.

    • @marcusmercer3208
      @marcusmercer3208 Год назад +2

      Something's definitely amiss with the Clipdrop version, which you would think would be setting a good example of how good SDXL is supposed to be. At present, it can't even render a spoon on a white background, in fact, nothing with a white background. MJ can do objects with white backgrounds in its sleep. According to SDXL, a spoon is a DSLR camera, a dessert spoon is a dessert with a camera in the middle of it. I also added 'vector style' yesterday and SDXL wouldn't render anything. MJ does all that with no problems.

    • @jopansmark
      @jopansmark Год назад

      Try to generate Xi Jinping in Midjourney (you can't)

  • @mattfx
    @mattfx Год назад

    @CoderXAI Great video, well done! And what we can use ComfyUI and Automatic 1111 on the same pc without problem?

    • @CoderXAI
      @CoderXAI  Год назад +1

      yep if one works the other should run as well. you'll run out of graphic card memory if you run both at the same time though! also ComfyUI is currently faster for most people

    • @mattfx
      @mattfx Год назад

      @@CoderXAI tks for the infos

  • @sxs8905
    @sxs8905 8 месяцев назад

    much helpful. Ty CoderX

  • @GambarAsliatauAI
    @GambarAsliatauAI Год назад

    so its basically a checkpoint i suppose ?

  • @rickysanchyz7083
    @rickysanchyz7083 Год назад

    ComfyUI works much better for me. With the same prompts it took Automatic1111 almost 30 minutes to generate a 1024x1024 image, but took only 8 seconds in ComfyUI !

  • @Kim-uu8fc
    @Kim-uu8fc Год назад

    No links in the description. (looks for another tutorial)

    • @CoderXAI
      @CoderXAI  Год назад +2

      it's the pinned comment, have some restrictions on adding links to the description for now

    • @Kim-uu8fc
      @Kim-uu8fc Год назад

      @@CoderXAI Oh ok, thank you for letting me know.

  • @umarudoma1811
    @umarudoma1811 Год назад

    can i use my SD1.5 lora on SDXL ?

  • @CJ-ur3fx
    @CJ-ur3fx Год назад

    Very helpful. Subbed.

  • @RyokoChanGamer
    @RyokoChanGamer Год назад

    refiner by camenduru > 404 - page not found

    • @CoderXAI
      @CoderXAI  Год назад +1

      thanks for letting me know. they've updated the file to refiner_v1.0.json and I've updated the link in my comment as well

    • @RyokoChanGamer
      @RyokoChanGamer Год назад

      @@CoderXAI 🫡

  • @Eleganttf2
    @Eleganttf2 Год назад +2

    Can SDXL 1.0 make NSFW models or images just like SD 1.5 ?

    • @cyfused
      @cyfused Год назад

      I do not think so. Provably it needs a lot of finetuning like 1.5 does.

    • @Elwaves2925
      @Elwaves2925 Год назад

      It isn't censored, like 2.1 is, so yes you can do NSFW but I don't know how well they'll compare to 1.5. Loras etc will help with that too once they start releasing.

    • @danielhernanalonso7219
      @danielhernanalonso7219 Год назад

      @@Elwaves2925 Uncensored doesnt mean trained. You cant do the same nsfw stuff as 1.5. Some nudes and thats it.

    • @Elwaves2925
      @Elwaves2925 Год назад

      @@danielhernanalonso7219 Obviously they don't mean the same thing but I also don't see the OP asking anything about training. Their comment is vague and can be read multiple ways, we both read it differently. Seeing as 2.1 was heavily censored, it made sense they were asking about that. 🙂

    • @danielhernanalonso7219
      @danielhernanalonso7219 Год назад

      @@Elwaves2925 thats true. I guess the answer is "yes, it can", but at the same time it can mislead him because he cant do the same nsfw stuff right now.

  • @chodon6868
    @chodon6868 8 месяцев назад

    It was very useful, thank you.

  • @goombagrenade
    @goombagrenade 9 месяцев назад

    This video was very helpful. Thank you so much.

  • @feflechi
    @feflechi Год назад +1

    Thanks to your instructions I got it running, but so far results are disappointing (havent tried the Refiner yet). It feels like starting all over again...

    • @TunesofIndia037
      @TunesofIndia037 Год назад

      Right, 0.9 was better in generating better quality images

  • @kevinehsani3358
    @kevinehsani3358 Год назад

    I only download refiner and not the base, should I be downloading both? Is there a documentation or tutorial on how to use comfyUI? Thanks

  • @kushagrapiano9036
    @kushagrapiano9036 2 месяца назад

    Does it run offline?

  • @chaitanyapahl
    @chaitanyapahl Год назад

    Is it gonna work on rtx 3050 laptop?

    • @innocentiuslacrim2290
      @innocentiuslacrim2290 Год назад

      No. Too little vram. Or maybe you can run it on CPU only and then you can generate pictures, even though it would be slow.

  • @thebonuslvl
    @thebonuslvl Год назад

    tx alot, get well soon!

  • @zikwin
    @zikwin Год назад

    for those who have trouble with python dependencies use this as a last resort to send all those dependencies to the void of darkness (fix my torch cuda xformers etc lol)
    for /F %P in ('py -3.10 -m pip freeze') do py -3.10 -m pip uninstall -y %P

  • @_Piers_
    @_Piers_ Год назад

    Thanks! :)

  • @throow
    @throow Год назад

    Thank you!

  • @3ricMO
    @3ricMO Год назад

    "Load this refiner.json file"?

    • @CoderXAI
      @CoderXAI  Год назад

      i've updated the link in the pinned comment to refiner_v1.0.json, please load that

  • @sergeyfedatsenka7201
    @sergeyfedatsenka7201 Год назад

    all good accept, Keanu has only 4 fingers ....

  • @LouisGedo
    @LouisGedo Год назад

    👋

  • @अघोर
    @अघोर Год назад

    ComfyUI is faster 😀

  • @pureone5692
    @pureone5692 Год назад +6

    I'll never buy anything with AMD on it...

    • @CoderXAI
      @CoderXAI  Год назад +1

      LULW, I think it might work on chonky AMD cards. Comfy has instructions for AMD+linux and auto1111 has some unofficial support but not sure if that works with SDXL as well.
      github.com/comfyanonymous/ComfyUI#amd-gpus-linux-only
      github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Install-and-Run-on-AMD-GPUs

    • @Eleganttf2
      @Eleganttf2 Год назад

      Indeed

    • @kano326
      @kano326 Год назад

      I am using RX 6800 on Linux its pretty well actually.

    • @Eleganttf2
      @Eleganttf2 Год назад

      @Cutieplus of course not , xformers uses CUDA

    • @pureone5692
      @pureone5692 Год назад

      @@kano326 ........linux