LICA: Layered Image Composition Annotations for Graphic Design Research
This paper introduces LICA, a large-scale dataset comprising over 1.5 million multi-layer graphic design compositions and animated layouts with rich structural metadata, aimed at advancing research in structured layout generation, design editing, and temporally-aware modeling by shifting focus from pixel-based to structure-aware understanding.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine you are looking at a beautiful, complex poster. To a regular person (or a standard computer camera), it's just a flat picture made of pixels. But to a graphic designer, that poster is actually a layered cake. There's the background frosting, a layer of fruit, a layer of text, and a decorative cherry on top. Each layer has its own rules: where it sits, what color it is, what font it uses, and even how it moves if it's a video.
For a long time, AI trying to understand graphic design has been like a blindfolded person touching that cake. It can feel the "shape" of the poster, but it can't tell the difference between the fruit and the frosting, or know that the cherry was added last. It sees a flat image, not the structure behind it.
Enter LICA.
The paper introduces LICA (Layered Image Composition Annotations), which is like giving the AI a pair of X-ray glasses and a detailed instruction manual for every single layer of the cake.
Here is the breakdown of what makes LICA special, using simple analogies:
1. The "Lego" vs. The "Painting"
Most old datasets were like handing an AI a finished painting and saying, "Here is a picture of a car." The AI sees the whole thing but doesn't know which part is the wheel, which is the door, or how they are connected.
LICA is like handing the AI a giant box of Lego bricks.
- It doesn't just give the AI the final picture; it gives them the blueprint.
- It tells the AI: "This is a text block (font: Arial, size: 20, color: Blue). This is an image (cropped in a circle). This is a group of shapes."
- It has 1.5 million of these blueprints. That's a massive library of "Lego instructions" covering everything from Instagram posts to business cards.
2. The "Time-Lapse" Camera
Until now, AI only knew how to design static pictures. But modern design is often moving (think of a TikTok ad or a PowerPoint slide that animates).
LICA is the first dataset to teach AI how to design "movies" made of shapes.
- Imagine a stop-motion animation. LICA doesn't just show the final frame; it tells the AI exactly how each piece moves.
- It says: "At 0.5 seconds, the title slides in from the left. At 1.0 seconds, the photo fades in."
- This is a huge leap because it treats time as a design tool, not just a sequence of pictures.
3. The "Recipe Book" with Variations
Have you ever seen a template where you can swap the photo and the text, but the design still looks perfect? That's a "template."
LICA is like a master chef's recipe book that shows you 24 different ways to make the same cake.
- It has nearly 1 million unique recipes (templates).
- For one recipe, it might show you 24 different versions: one with a red theme, one with a blue theme, one with a different photo, but all using the same "logic" or structure.
- This helps AI learn the rules of good design, not just copy-paste images.
4. Why Does This Matter? (The Superpowers)
By giving AI this deep, structured understanding, LICA unlocks some cool new abilities:
- The "Magic Eraser" that actually works: If you want to remove a logo from a poster, old AI might just paint over it with a blurry mess. LICA-trained AI knows exactly which "layer" the logo is on, so it can remove it cleanly without messing up the background text.
- The "Style Chameleon": You can tell the AI, "Make this business card look like a music festival poster," and it will understand how to swap the fonts, colors, and layout logic while keeping the structure intact.
- The "Motion Director": It can take a static flyer and automatically turn it into a cool, animated video ad, knowing exactly when each element should pop up.
The Bottom Line
Before LICA, AI was trying to learn graphic design by staring at finished paintings. It was guessing.
LICA is the first time we've handed AI the actual tools, the blueprints, and the instruction manuals. It's a massive step forward that allows computers to stop just "seeing" designs and start actually understanding and creating them with the same structural logic that human designers use. It's the difference between a robot that can draw a stick figure and a robot that can design a professional magazine cover.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.