Image size & AI Upscale

Advanced options menu

Hit the advanced button on the bottom left to reveal the aspect ratio, guidance and steps options.

Here’s what each of these do:

The default image proportions are Square, and for a good reason. Our AI was trained on square images. Going too wide or two tall without describing other things in the image will sometimes get you twins, so square is a good place to start. You can change this to tall, wide, or landscape.
Prompt Guidance: The lower the number, the more freedom the AI has to interpret what you mean, but things can get blurry. If you’re getting blue squares, stepping down in guidance usually makes unwanted artifacts go away. The higher that number goes, the more assertive the AI is about what you want, resulting in a sharper image. You can now understand why we hid this in the advanced menu. This is a delicate setting for people with the patience to master it.
When you think about Steps, think about how your printer works. The more passes it does, the better the quality is. But if your printer does too many passes on too many instructions, things end up blotchy. So while increasing the step count can give you a more detailed image, writing a clear prompt at 30 steps can also get you there with less wait time.

Tons of tools are available after the image is generated:

HighDef doubles the pixels of the image, similar to anti-aliasing
Facelift 4x’s the pixels and also applies a face restoration, similar to the beauty mode on your phone
More will give you five more variations of the same picture, with minor changes
Remix lets you re-prompt into another base model, destroying the image and making a new one
Edit is a chat editing program, where you can ask questions like “what if her hair was green?”. Important: Edit is experimental and only for low-resolution use, it uses its own AI model and doesn’t work within the concepts system
Inpaint is generative fill, you can literally paint on the photo with your finger or mouse and re-prompt into the mask