The single most common mistake in AI image upscaling is using the wrong preset for the content type.

It's an easy mistake to make — upscaling is upscaling, right? The AI makes the image bigger. Why would the content type matter?

It matters because each upscaling preset is a separately trained model, optimized for a specific category of image content. These models learn different things. A model trained on photographs learns photo texture — skin, fabric, foliage, natural materials. When you apply that model to an illustration, it doesn't know what to do with flat colors and clean linework, so it invents texture that shouldn't be there. The result looks worse than bicubic upscaling.

The four main content categories

Photographs

This is the most general-purpose preset and covers the widest range of content: portraits, landscapes, product photography, architecture, food, anything captured by a real camera.

Photo models are trained to reconstruct texture-rich detail — the things cameras actually capture. They expect areas of the image to contain fine texture (even skin that looks smooth still has texture at high magnification) and they reconstruct accordingly.

Use for: Real-world photography of any subject. AI-generated photorealistic images. Stock photography.

Don't use for: Illustrations, cartoons, anime, screenshots, UI captures.

Art & Illustration

Illustrated content has fundamentally different characteristics than photography: flat areas of color, painted brushstrokes, deliberate stylization. An art model is trained to preserve these qualities without adding artificial photographic texture.

The key difference is edge behavior. In illustration, edges between color areas are intentional and should stay clean. A photo model applied to illustration will add texture along those edges, softening what should be a crisp line and creating a noise pattern that looks like film grain on a cartoon.

Use for: Digital illustrations, painted artwork, vector art exported to raster, concept art.

Don't use for: Photography, anime-style art (use the dedicated anime preset).

Anime & Cartoons

Anime has its own visual conventions: flat fills, very specific line weights, particular color palettes, and stylized detail that differs from both Western illustration and photography. Anime models are trained on this specific aesthetic.

The practical difference from the general art preset: anime models understand clean lines and flat color areas better, and handle the specific compression artifacts common in anime source material (which is often distributed as highly compressed video frames).

Use for: Anime screenshots, manga scans, cel-shaded illustrations, cartoon artwork.

Don't use for: Photorealistic content, non-anime illustrations.

Text & Documents

Text presents a specific upscaling challenge: it needs to be readable. The priority is edge sharpness and character legibility, not texture reconstruction. A text model is optimized to produce clean, sharp character edges rather than natural-looking texture.

Use for: Screenshots with UI text, document scans, presentation slides, anything where text legibility is the primary concern.

Don't use for: Content where texture matters more than edge sharpness.

The decision tree

1. Does the image contain primarily real-world photography? → Photo 2. Is it clearly anime or cartoon in style? → Anime/Cartoons 3. Is the main concern text legibility? → Text & Documents 4. Is it illustration or painted artwork that isn't anime? → Art & Illustration 5. Is it an AI-generated image? → Look at the style: photorealistic → Photo, illustrated → Art

When you're not sure

Try the Photo preset first — it's the most versatile and produces acceptable results across the widest range of content types. Then compare with the Art preset if you're working with illustrated content.

The quality difference between the right and wrong preset is significant enough to check. A 30-second comparison in Upscale Forge before finalizing your upscale saves you from delivering work that used the wrong model.