How can I upscale my images with AI?

Upscaling your image, improving faces

Pirate Diffusion has built-in free AI image upscaling to 4X the pixels of your images. This is a free service built into the system.

The /facelift command is a general upscaler. It can improve photos that don’t have faces, too, like an object or a landscape.

There are also two other modes+

/facelift /anime — for any kind of illustrated works, pass the /anime secondary command. It works great on drawings, paintings, CGI, including semi-realistic CGI.

/facelift /photo — when you want upscaling but not face retouching, such as landscape photo, or when you do not want your character to have that “airbrushed” look

EXAMPLE 1

This one’s very easy. Here I have my five /brew results, and now I am selecting the one I want to upscale.  Simply (right click) or (long touch) the image, click Reply as if you’re going to talk to the photo, as shown below.  

And then type the words /facelift in the reply box and you’re done.

There are some tradeoffs

Just like turning on “beauty face” on your smart phone camera, facelift uses similar technologies to smooth fine details. This can sometimes result in overly smooth faces or loss of tiny details.  But more than often, the results are really impressive.

Typical render:

Upscaled image:

Learn how to make images like this one

EXAMPLE 2: QUICK RENDER

Then use /facelift to AI upscale the pixels 4x, which also smoothens lines


And upscaled once to x1024

And upscaled again, 8x the original image

Greater detail is possible by starting at a higher resolution, and making the prompt more specific about skin imperfections, shadows, and so on.

a photo of young kate bush in a bikini, 21 years old

Learn how to make images like this one

Aspect Ratios

AI models “learn” using square 1:1 images, so when we ask for a picture back that isn’t a square, this can result in glitches like multiple heads on a person, crashes, causing images to render poorly or not at all. Yikes.

To account for this, we created shortcuts that render the most compatible dimensions without burdening you to remember the exact pixel numbers.

Use these shortcut commands for best compatibility with all AI models

Examples

/render /wide a man in front of a burning building

/render /tall a man in front of a burning building

Deprecated: /size – You can still manually use /render /size:512×512, but this may result in unusual behavior.


Processing “steps”

/steps refers to the number of render passes that the server takes to arrive at the final image. More steps can mean a more detailed image, while fewer steps can be considered a rough sketch, though in some cases less it can result in an “overcooked” result, especially when drawing line art like Anime.

Less steps takes less processing time, so we’re able to send you back more images in roughly the same amount of time.

/render /steps:wayless — returns 9 images

/render /steps:less — returns 6 images

/render — returns 5 images (default 50 steps)

/render /steps:more — returns 3 images

/render /steps:waymore — returns 2 images  (experimental, upper limit)

The output of photos and actual step count may change in public servers based on dynamic use, however can be defined in stricter terms for premium server clients.

As such, the ideal workflow begins with low steps and later re-requests with a higher quality steps command. To keep continuity, mix this with Seed. 

Recommended:

pirate diffusion logo

PIRATE DIFFUSION’S BLOG & TUTORIALS
Easy mode with Brew

Positive/Negative prompts

> AI upscale your images

upscale tutorial part 2

Fine-tune with Render

Upload and edit images

Change visual styles

Make realistic AI people

Commands & troubleshooting

There may be other limitations that vary on a case by case basis, such as noise, which varies by prompt and AI model. We recommend picking up a copy of Affinity Photo (free) or Photoshop to polish off any minor blips.