General Pattern for Recognizing a Decorative Image
⚠️ Not currently in production. This checkpoint is not part of the live product right now: the image classification models it relied on were discontinued. Integration is planned for an upcoming release.
An image is decorative when its presence does not change the understanding of the content. The image accompanies visually, but does not add additional meaning.
If removing the image does not affect what the user understands → it is decorative.
Flexible Patterns (Non-strict, interpretable by AI)
Pattern 1 — Image that only embellishes or decorates
The image is there to make the page more attractive, not to explain something.
Variable examples:
- Background photos
- Textures
- Ornamental borders
- Ambient images in banners
Idea for the AI: If the image does not modify the meaning, only the perception → decorative.
Pattern 2 — The information is already fully expressed in the text
If the nearby text already communicates what the image would show, the image adds no content.
Variable example:
- Photo of a sleeping dog, and the text says: "A dog sleeps peacefully."
- PDF icon when the link says: "Download report (PDF)"
Idea: If the text covers everything, the image is decorative.
Pattern 3 — Repetitive or style-based image
When the same icon or graphic is repeated in multiple places without changing the meaning.
Examples:
- Arrows in lists
- Uniform icons in menus
- Visual separators
Idea: If the image does not differentiate elements, but adds aesthetics → decorative.
Pattern 4 — Image inside buttons or links that already have clear meaning
When a button or link has text and the image only visually reinforces the action.
Example:
<button>Save <img src="icon-save.svg"></button>The meaning is already given by the text "Save". Without the image, nothing changes.
Idea: If the image does not change the action, it is decorative.
But pay attention (important flexibility):
The same image can be decorative or not, depending on the text and context.
| Image | Nearby text | Result |
|---|---|---|
| PDF icon | "Download report" | Not decorative → adds format → informative |
| PDF icon | "Download report (PDF)" | Decorative → the text already says it |
The AI does not look at the image in isolation. The AI looks at the relationship between image + text + intent.
So, what the AI learns is:
An image is decorative if its absence does NOT cause a loss of meaning.
The decision depends on the nearby text and purpose — not just the image.Ready-to-use version as a knowledge block in the prompt
Consider an image "decorative" when its presence does not add meaning to the content.
The image is decorative if:
- The nearby text already expresses all the information the image could convey.
- The image is used to embellish, set the mood, or visually reinforce without changing the message.
- It repeats uniformly as part of the design.
- It is inside a button or link with clear text, and the action is fully understood without the image.
The image is no longer decorative if:
- Its absence changes what the user understands.
- It identifies something, differentiates elements, or adds clarity.
- It is the only way to understand an action or category (in that case it may be active or informative).