(This template is used for Workflow Contest)

What this workflow does

👉 My first goal was  to recreate an image (512x512), initially generated with a 1.5 model, using an SDXL model. No textual prompt is needed, only the IPAdapter serves as a guide.

"Some sort of creative upscale from old pictures with new models."

No need to tweak parameters (except weight and noise of IPAdapter, if you want). Just load a low-res picture, no prompt needed.

An easy demo for IP_adapter power and another lazy way to compare SDXL models.

Of course, you can use SD 1.5 models, if you change the IPadapter model and Latent size.

If you want to prompt positive and negative text, just delete the conditioningZeroOut, or bypass this nodes. You need to change the weight (by default, set to 1) of the IPadapter if you want your text prompt getting attention.

How to use this workflow

👉 Just add a image (Load Image) and press queue. You can drag anything from  web or your HD.

Choose your favorite model. That's all folks :)

Tips about this workflow

👉 You can play with IPadapter weight and noise for creativity.

By default, seed is fixed, don't forget to randomize if you want batch some images.

🎥 Video demo link (optional)

👉 [Please add here]