Home Artists Posts Import Register

Content

Hi all, my cleaning made the computer alive again, horray.

Even so, the computer is still kinda of funny, slowing down at random.

I'm afraid that it may not turn on tomorrow, so I gave it my best to complete the build today, because of that i'm ashamed to say that I was not able to test all the options as well as I would like it, the computer seen to be sweating to load the model each time.

I was able to test image generation and a few other stuff, but this is a big update and usually big updates have big bugs, just hope there is no bugs that will break a night of rendering.

Here is a few bugs I found that didn't had the time to fix:

  • If you have "Preview images" turned on and use upscale, the preview color will look wrong (but the file will save correctly)
  • Using Face correction + "Save memory" will not unload the face model after each render. You may get a OutOfMemory error on the second image, since the face model will still be loaded.
  • Stop Render and Skip render are experimental and I could not test then a lot, it may bug.
  • Ah, another bug. Img2Img strength is inverted again, sorry =/ 90% strength in GUI = 10% in the model

I still need to properly try to load other models on the GUI (it may explode the pc) so i'm uploading it before giving it a go. If it work, will make a tutorial later on how to add models to it.

Here hoping that at least the build will open and render stuff lol

Side note: The patron list on the GUI is updated on the second day of each month.

Files

Stable Diffusion GRisk GUI 0.5.rar

Comments

Anonymous

Love it so far! You've been such an amazing soul. I have one suggestion tho, it is to maybe add two separate prompt input fields. One is for the main prompt request and the second one would be all the suffix that the user regularly uses. For example on the 1st text field I input "alien landscape on mars" and the 2nd text field would be "digital art, intricate, ray tracing, 3D art, ...etc" then you could join both fields with a comma in between and that would be your final prompt for SD to read. I have found myself manually and tediously selecting the first sentence each time I want to change it, but keeping the rest. Having a separate text field would be a time saver. Maybe a feature you never knew was needed but ended being super useful. No one has implemented such a thing. Do you think it's possible? It's just a suggestion.

Anonymous

Thanks so much for this! Up until I downloaded this version (today), I'd been using 0.1 - which I was very grateful. With this download, the Batch Samples have been AMAZING (I'm used to Midjourney providing me a up to 4 variations). This gives me SO much more bang for my buck. I am basically cranking out 6 images for every seed (have a 3090) - huge time saver. The interface additions have been great for quality of like. Would be pretty awesome if you added some Image Modifiers, like style, but I'm sure you have lots on your plate and lots in store. Thanks again!

Anonymous

2 things after lots of testing. 1. the image preview box is more 'glitched' than before. Images now turn inverted colors. Current generations are correct colors, once esrgan kicks in it 'final' turns inverted teal and yellow. Still either squished or out of order lines outside of 1:1 ratios. 2. Upscale photo runs more slowly than before. For some reason when the image is upscaled it takes about 5x longer, showing up in about 5 'scanline' progressive chunks from top to bottom, instead of all at once. Sometimes it also writes a black image.

Anonymous

Display during rendering doesn't work as expected. If I use 25 steps, it's like step5 renders step1 and step20 renders step2. And if I increase the steps to 100, step25 renders step 1 and step 75 renders step2. All I see is static no matter how many steps I use.

Anonymous

Is there any possibility of using HandGAN (H-GAN) to fix the shape of hands in the output? BTW I think the face fix option in the upscaler is wonderful!

Anonymous

put a model from hugging face and i get this error: {'vae', 'tokenizer', 'feature_extractor', 'text_encoder'} was not found in config. Values will be initialized to default values. Traceback (most recent call last): File "start.py", line 699, in OnRender File "start.py", line 636, in LoadModel File "diffusers\pipeline_utils.py", line 273, in from_pretrained TypeError: __init__() missing 4 required positional arguments: 'vae', 'text_encoder', 'tokenizer', and 'feature_extractor' what does this mean? is the model incorrect or is there more extra steps i need to take to get it to work?

Anonymous

Could I request https://www.reddit.com/r/StableDiffusion/comments/xgzxjl/txt2mask_was_just_released_we_dont_have_to_use this? Seems really cool not needing to paintbrush for inpainting

fructchompler79

I still can't download, hitting a quota error page. Is there anything I can do about this, or do I just need to wait further?

DAINAPP

Will start uploading on other host. I think the grow on Patren really took a toll on my uploads.

DAINAPP

I gonna need to take a better look to undestand what it does, for now gonna add to my TODO list

DAINAPP

It seen like a incorrect model. Can you send me what model did you use?

DAINAPP

Can do, it may take a few updates tho. There is more important features for now. Will add to my TODO list

DAINAPP

Glad you are enjoying it. Don't worry I plan to keep updating it until its a badass GUI

Anonymous

https://huggingface.co/justinpinkney/pokemon-stable-diffusion

Anonymous

those six are the ones that didnt work out of the 10 i downloaded through the tortoise method

Anonymous

from what i can tell they dont have folders named 'vae', 'text_encoder', 'tokenizer', and 'feature_extractor' so i tried adding them empty and from stable 1.4 but they didnt work out messed around with the json files too got bunch of errors so i assume i would have to do some python to get it to work which im not confident in doing

DAINAPP

Sorry for the deleted posts, hit enter by mistake: justinpinkney/pokemon-stable-diffusion This was saved in another way. Would need to find the original code to make it work. ny7777/ddpm-pokemon-128 mrm8488/ddpm-ema-anime-v2-128 and others Seen to be missing files? Maybe throwing on top of the fp16 model, but not sure it would work jags/fractaldiffusion This one is clean, only 2kb files.

Anonymous

k will try to see if throwing in folders from fp16 model will make it work our not. the one that is clean what does that mean? is it missing a lot of stuff would throwing in folders from fp16 do anything?

Anonymous

justinpinkney is a lost cause unless find the original code? ok idk if ill be able to find it but will look. edit: just looked and justinpinkney does have this https://huggingface.co/justinpinkney/sd-pokemon-diffusers1/tree/main which has all the folders so this should work not sure if it the original code though

Anonymous

I got the wrong message, i don't know why Exception ignored in thread started by: Traceback (most recent call last): File "start.py", line 281, in Renderer UnicodeEncodeError: 'cp950' codec can't encode character '\xe9' in position 20: illegal multibyte sequence

DAINAPP

You should have a weird character in the prompt, maybe some Chinese character or a invisible char. Need to fix that.

Anonymous

I am at the all access level level Patron. I just signed on to your patreon a couple days ago and I have your 1.0 version and it sleeping bag is working well. I'd like to download the updates but it won't download . When you say that the patron list updates on the 2nd day of the month does that mean that I have to wait 2nd of October to update my interface? I tried in google Chrome and then again on Microsoft Edge and was unsuccessful in both. Any insight into my problem would be a super well appreciated. Thanks

DAINAPP

Sure is possible, I have something similar in my TODO list, I think they call it a prompt matrix or something? If your prompt is "An aline in mars, #digital art#" it will generate a image with digital art and one without.

DAINAPP

Hi Chis, the download should already be working for you, send me a private message that I can help you. 2nd day is just for me to update the Patrons list inside the app.

Anonymous

Mu suggestion was to mix both text fields into one prompt. So one final image in the end. Kinda using the XYZ formula to write the best possible prompt.

DAINAPP

Yeah, is more or less my idea as well, is just a little hard to explain here exactly what I want to do.

Anonymous

Now worries! I'm glad you were thinking of implementing it as well. 😊

Anonymous

ok thank you, i will try it, and BTW, when I open 0.5 app, I got a error message "Failed to load image Python extension", but app just can work normally, can I ignore it?