I wrote a GUI for your project.

[0xBLCKLPTN]

The GUI is written in PyQt5. It is as simple as possible and so far only works with the scripts/txt2img.py

Gui supports parameters such as:

• ddim_steps INT
• seed INT
• prompt STR
• height INT
• width INT
• skip-grid (DEFAULT)
• n_samples 1 (DEFAULT)
• output

Also, the interface shows the result of generation instead of the default ricroll. Everything is working fine. Tested on Windows 11, MacOS BigSur. Does it make sense to develop this interface in any way?

https://github.com/blcklptn/CompVis-StableDiffusion-Gui

[fat-tire]

This is great-- I made a couple of changes/fixes for Linux and generally-speaking. Hopefully didn't break anything for Windows/Mac along the way, but hopefully will be a little more flexible. Let me know- this makes fooling around a lot easier and more fun.

Looks like this now:

Oh, I had a list of things to do and maybe adding a "copy result to clipboard" button would be another nice thing to add:

• Don't block the UI while running
• Redirect output to a status screen
• Add a "random seed every time" checkbox so it's automatic
• Delete/clean up old images?
• Save settings somewhere?
• Add "Copy to Clipboard" button

Again, great job!

[0xBLCKLPTN]

@fat-tire Thanks! I too think that adding the result to the buffer and saving the settings is a good idea. I will add these changes later ;)

[fat-tire]

Cool I did a few more changes in this PR.

Now looks like this:

*thanks to @Fictiverse for inspiring the prompt on Twitter

[fat-tire]

Ideas for updated version--

• Support auto-generating multiple images in one-shot (variable # of images)
• If the above is chosen, maybe use the grid/thumbnail in the result.
• A button to open the "outputs" folder containing the most recent images in the Finder
• A delete all images button (with a confirmation popup)
• Maybe support some more of the arguments

[fat-tire]

For anyone following along, here are my Day 3 contributions to the GUI thing.

Highlights:

• Support generation of multiple images - still in batches of 1
• Slight UI changes
• Default to 512x512 again
• Minor tweaks & fixes

[leszekhanusz]

Because competition is always good, here is my shameless plug for my web interface in Vue to do inpainting. It looks like this. Still a work in progress of course...

[firstmover]

Hi, this model is wonderful. I also wrote a web UI with very minimum code.

If you are interested in this UI, the code can be found here.

[fat-tire]

Glad to see more UIs! @firstmovier, if you all don't know there's also a web-UI in progress at the lstein fork of this project.

I just submitted a new commit-- adding a scrollable area for the image grid in case there are so many that it overwhelmes the window size.

I think I'll continue development on this to try to incorporate upcoming support for the json image metadata that's being formulated over at lstein.. I think it would be nice to see it supported on multiple interfaces so json image descriptions can be sent back and forth...

Also, @blcklptn --maybe add a link to your repository in the top post?

[fat-tire]

New today on the same PR-

img2img support!

(a pulldown at the top lets you choose from txt2img and img2img, and you can choose the image to use via the button...)

[fat-tire]

Another update-- those separator bars between the terminal and the image and also separating the left and right portions of the screen are now movable so that you can reveal or hide more of the images.

Enjoy. Latest version is in the pull request or you can find it on my fork.

[fat-tire]

Updated w/a stupid int/float bug for strength and use a slider now -- PR here. Or see main branch on my fork.

[breadthe]

Hah look at that! I made one too in Tauri + Svelte. Free + open source. It too uses the txt2img.py script. You do have to build it from source for your platform though.

https://github.com/breadthe/sd-buddy

[fat-tire]

Quick note to anyone following that I added a few new bits here.

The coolest one is you can drag the image around by hand now. You even get a little "dragging hand" icon.

Also added some tooltips and instructions on how to install pyqt5 if you don't have it already in place.

Enjoy!

[leszekhanusz]

Update for DiffusionUI: I added image to image including doing image to image inside an inpainting region

I made a short video to demo the process: https://www.youtube.com/watch?v=AFZvW5qURes

[fat-tire]

AIRIM.webm

I did some changes too-- this is a test of using json-based metadata to recreate an image.