Stable Cascade to SXDL with Refiners, LoRA stack & Upscale

-----------------

The workflow is very messy, but wanted to share. feel free to clean up after me.

SD Cascade produces the base image which is used for prompt comprehension and composition of the image. Then the first preview is displayed.

Then the Image runs through SDXL and a refiner, and it displays the second preview. (There will be increased fidelity, but should retain the main composition and style of the original image depending on the SDXL checkpoints you use.)

The last step runs the image through a 2x ESRGAN upscale which is not the most modern approach, but produces good results quickly, because the result of that is run through a second SDXL refiner, which fixes any artifacts from the upscale and adds detail.

Finally, the Final result is displayed.

Enjoy!

(remember, for the cascade empty latent, use 32 for compression and change resolution to 1024x1024 or other SD compatible resolution before using)

V3 NOTES:

• added lora stack node.

• moved all important nodes for prompting to the left.

• replaced old ESRGANx2 upscaler with ultimate SD upscaler.

• disconnected lora info node for reference
  
  (also added some more outputs for your viewing pleasure.)
  

WARNING:

for those of you having an unexplained error associated with CLIP, it may be because the model.safetensors file that comfyworkflows links on this page is the incorrect file. It is likely for this to happen if you have not previously run Cascade before this workflow. You can find the correct one here: https://huggingface.co/stabilityai/stable-cascade/blob/main/text_encoder/model.safetensors