This workflow is for this video tutorial here: https://youtu.be/_t6USD-dX2g

Go watch it for more infos.

Support me on ko-fi:  https://ko-fi.com/risunobushi

It uses either a LCM, Hyper or Lighting model to create real time generations based on Canva, Photoshop, Illustrator, or whatever really, while the Screen Share node takes care of sourcing the original canvas in a img2img process based on Depth.

It also uses a LLM to automate prompt writing based on what Ollama sees.

Settings for LCM (epicPhotogasm):

- 8 Steps

- CFG 1

- scheduler: sgm_uniform

Settings for Hyper (Juggernaut):

- 6 Steps

- CFG 2

- scheduler: sgm_uniform

Settings for Lighting (Juggernaut):

- 8 Steps

- CFG 1.8

- scheduler: sgm_uniform

Settings for refining stage (epicRealism / Juggernaut v9):

- 20-40 Steps

- CFG 6-8

- scheduler: karras

I strongly suggest Xinsir Depth for XL: https://huggingface.co/xinsir/controlnet-depth-sdxl-1.0

Set your Screen Share node window and area to your canvas, and have fun!