WheresWaldoStyle

Hello! This is a 14,000 step trained model based on the famous Where's Waldo / Where's Wally art style. (I'm American so I named the style Waldo, if you're familiar with Wally instead, my apologies!)

The keyword to invoke the style is "Wheres Waldo style" and I've found it works best when you use it in conjunction with real world locations if you want to ground it at least a little bit in reality. If you really want the Wally/Waldo look, be sure to include "Bright primary colors" in your prompt, and add things like "pastel colors" and "washed out colors" to your negative prompts. You can also control the amount of "Waldo-ness" by de-emphasizing the style in your prompt.

For example,

"(Wheres Waldo Style:1.0), A busy street in New York City, (bright primary colors:1.2)" results in the following image:

While

"(Wheres Waldo Style:0.6), A busy street in New York City, (bright primary colors:1.2) brings in some details about New York city like the subway entrance that you won't find at full strength style.

One thing to keep in mind, if you try to just spit out a 2048 x 2048 image, it's not going to give you waldo, it's going to give you a monstrosity like this:

I've found the sweet spot for this model to be in the 512 x 512 to about a max of 640 x 960. Much beyond that and it starts to create big blobs like the example above. It does take pretty well to inpainting though, so if you create something interesting at 640 x 960, throw it in inpaint and start drawing in fun details (you may have to reeeeally de-emphasize the style in your inpaints to get it to give you what you want, just a heads up)

Finally, one thing I've found that really helps give it the "waldo look" is using Aesthetics. I like to run an Aesthetics pass at a strength of .20 for 30 steps. It helps prevent the really washed out colors and adds the stripes that are so prevalent in Wally/Waldo comics. I've included the Waldo2.pt file if you want to download it and use it, it was trained on the same high quality images I used for the checkpoint dreambooth training.

Here is a screenshot of my config I use for generating these images:

I hope you have fun with this! Sadly it won't actually generate a Waldo/Wally into the image (at least not one that you can generate on demand), but if you're going to all the trouble to inpaint a proper Waldo/Wally scene, you can do some quick post work to add Waldo/Wally in there somewhere! =)

Here are sample images using this model: