Open3DLab/open3dlab_slopscaling
1
0
Fork 0
Code Activity

Add slopscaling workflow and basic instructions.

Browse Source
This commit is contained in:
Ganonmaster
2026-04-05 01:32:24 +02:00
parent 42bd916dfb
commit 60bd2293a1
2 changed files with 1486 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
README.md
+17
View File
@@ -1,3 +1,20 @@
# Open3DLab SlopScaling (OLSS) 5.0
This repository contains several components used to make Open3DLab's slopscaling April Fools prank work.
## ComfyUI Workflow
In the `workflow` folder, you will find the ComfyUI workflow that was used to mangle all the thumbnail images. This is an extremely simple, and extremely bad workflow that should under no circumstances be used for anything serious. However, you can theoreticaly run this on any device that has sufficient VRAM to run SDXL.
### Run it locally
To run it locally, you will need:
- Graphics hardware with sufficient VRAM
- ![ComfyUI](https://www.comfy.org/) installed on said hardware
- ![ComfyUI-Florence2](https://github.com/kijai/ComfyUI-Florence2) (can be installed through ComfyUI Manager)
- the ![Juggernaut XL checkpoint](https://civitai.com/models/133005/juggernaut-xl) (specifically, Ragnarok_by_RunDiffusion)
Once you have all your dependencies installed, and the checkpoint in the right directory, open the Yassification nodegroup, and set the model on the `Load Checkpoint` node. Plug in an image in the `Load Image` node, and hit run.
Your inference times will vary. On a Macbook M1 Pro, inference time was roughly 30 seconds per image. On an RTX4090, it is 2-3 seconds. On an Intel Arc Pro B50, it was roughly 15 seconds. It was not tested on cloud instances.