Skip to main content
Agentic Architecture Skills.
Start hereUse casesArchitectureIntelligence ladderKnowledge & memoryResearchSkills
Optional: types of data

Image embeddings: search by visual features

Find related inspection images while keeping the original observation available.

By Murali SidLinkedInPublished Updated

A model can place visually related inspection images near each other, making it possible to look for earlier examples of corrosion or damage. With a suitably aligned text-image model, a written query can also find images. What counts as similar depends on training.

Two inspection photographs of corroded pipe joints map to nearby points while clean-valve images form a different group. An aligned text-image model can connect a phrase to visual results. Each result links back to the original image.
Images: find visual resemblanceInspect the original image, date, and asset. Visual resemblance is not a diagnosis.Open image at full size ↗

What the space represents

The image illustrates a model placing related visual features near each other. It may be useful for finding earlier inspection images that resemble a new observation. The useful features depend on training; a model may attend to background or camera style when the task needs fine surface damage.

An aligned text-image model can connect a phrase to images. CLIP provides a research example. This does not mean arbitrary text and image encoders share a comparable space.

What to retain beside the vector

Keep the original image, asset identity, capture time, camera context, image region if one is used, source reference, permissions, and model version. A result should open the image that produced it.

Where the first approach can fail

A generated caption can omit a small crack. Visually similar corrosion can occur on different components with different consequences. Darkness, occlusion, and viewpoint can affect what is visible. Evaluate actual inspection conditions, including cases that must not be treated as matches.

How it serves a decision

Use resemblance to find useful comparisons. Have the decision rely on the inspected image and the applicable engineering criteria, with uncertainty recorded. Combine images with other evidence and preserve media event context.

External source

CLIP: connecting text and images, 2021. Reviewed 18 September 2026.

Continue the story: Search video for events.

Cite this page

Written by Murali Sid, Founder and AI Systems Architect, last updated 15 October 2026. Licensed CC BY-SA 4.0, so quote it freely with attribution.

Murali Sid (2026). Image embeddings: search by visual features. Agentic Architecture Skills. https://www.agenticarchitectureskills.com/memory/image-embeddings

Machine-readable version of this page: /memory/image-embeddings.md

Text embeddings: search by meaning

Make differently worded passages discoverable while preserving exact evidence.

Video embeddings: search for events

Represent a useful interval while retaining timing and surrounding context.

On this page

What the space representsWhat to retain beside the vectorWhere the first approach can failHow it serves a decisionExternal source