This ComfyUI Workflow combines a base generation using SD1.5 models, LoRAs and embeddings, then runs a second pass and an upscale pass with SDXL Models, LoRAs and embeddings. This allows for the versatility of SD1.5 while giving SDXL quality outputs.
I use comfyUI for five months now and this is my solution to get my previous workflows in SD1.5 into SDXL quality.
This workflow uses the following custom nodes:
BlenderNeko: Advanced CLIP Text Encode
If you find this helpful, consider buying me a coffee:
Description
FAQ
Comments (7)
since LORA for SD 1.5 are not compatible with SD XL, there is a utility to convert them?
I mean: the library or personalized LOra is the great strong point of SD against other system, now we must trash all of them?
no, that's the point exactly. This workflow uses two model checkpoint files, one 1.5 and one XL.
The base image is generated using SD1.5, so all older LoRA can be used on that image. If you have older workflows with several LoRA and finetuned prompts etc. you can replicate them in the first half of the workflow.
Then on the refiner and upscale passes, SDXL is used to get the crisp quality upgrade it brings.
You can tune the Denoise for XL how much the original will be overwritten, so XL LoRAs can also be applied.
hope that clarifies the intended use
Sounds interesting! That can also be done by just changing the checkpoint and LoRa loaders input of course. If you want to share some results of the other way around generated images, I would be happy.
wow
it's still useful for old things it was very useful for me because I had no idea how to make a workflow in sd15. 👍
I'm also still running a version of this workflow to this day, to utilize versatility of 1.5 and upscale it with SDXL