Daytree

Uses Stable Diffusion to slowly grow a tree, day by day¹ (and then automatically sets it as a desktop background).

The tree slowly grows over time (using ControlNet), and adapts to the current (real-world) season. More stable generations could be achieved by using a fixed seed (as well as skipping step 2 in the process outlined below). However, this project prioritises creating exciting images instead. All StableDiffusion generation is done through the use of the Automatic1111² API.

Today's Tree:

Previous trees can be seen in the output_imgs folder.

Generation Process Overview:

1. Generate a 512x512 image using img2img and ControlNet, with the previous iteration's image as the input image (or, if it's the first day, a simple placeholder). The ControlNet image is created anew each day to slightly "grow" the tree.
   
2. Use img2img to upscale to (912x512)*[image_scale]. This step has two functions. The first is to create an image that matches the desktop resolution. The second is to add more variety to the output: when using img2img to non-square resolutions, the result often has unpredictable elements. This can result in strange artifacts and less aesthetically appealing images; however, they are also more exiting and fun (often small travellers will appear, or multiple trees, etc.)
   
3. Use R-ESRGAN 4x+ to upscale to 2k.
   
4. Apply a blurring and dimming to the left and right sides of the image (which, from step 2, are usually just vague lines of colour).
   

Footnotes

1. Provided that the workstation isn't fully utilized that day. ↩

2. https://github.com/AUTOMATIC1111/stable-diffusion-webui ↩