Ola can find a photo by describing what is in it. Upload a picture of a red ankara bag on a wooden table, then ask her for the bag on the wooden table, and she can return it even when the file is called IMG_4821.jpg. The search runs on descriptions that Salesive generates for every image and video you upload.

What does search by content actually mean?

Most media libraries search the filename and the alt text you typed. That works only if someone named the file carefully at upload time. In a real store, files arrive from a phone camera, a designer, a staff member and a supplier, and the names are a mess. A content search ignores the name and looks at what the image shows.

So you can ask for 'the picture with the yellow background', 'the video of the shop front' or 'the photo of the model in the blue dress' and get a result without knowing what the file was called. The search is only as good as the description attached to the file, which is why the next section matters.

How is the description made?

When you upload an image or a video, the jobs service sends it through an AI model that describes the content and attaches keywords. The primary model is Workers AI GLM 5.3 Flash. OpenRouter and Mistral Small 3.1 act as fallbacks, so a file is still described when the first service is unavailable.

The result is a short text description plus keywords, stored with the file. That text is what Ola searches. The model is not reading your product catalogue or your captions, so the description is based on the image itself, which is exactly what you want when the filename tells you nothing.

Where can you see the descriptions?

The Media page lists your files and their descriptions. That page is the honest view of what the AI saw. If a search is not finding the photo you expect, open the Media page and read the description. You will usually find that the model described the object differently from the words you used.

For example, you may think of a product as a 'tote bag' while the description says 'handbag'. Searching for 'handbag' will find it; searching for 'tote' may not. Reading the description once saves a lot of guessing, and it also tells you when a photo is genuinely too unclear to describe well.

How do you ask Ola for a file?

Ask the way you would ask a person who has seen the photo. Mention the subject, the colour, the setting and anything distinctive: 'the photo of the red bag on the wooden table', 'the video with the woman speaking to camera in the shop', 'the image of the blue dress against the yellow wall'. Specific beats short.

If the first answer is wrong, add a detail rather than repeating the same request. The description is fixed at upload time, so a different phrasing that matches the description is more likely to work. You can also ask her to list what she found so you can pick the right file yourself.

A good habit is to ask for a few results rather than one. If the library holds several similar photos, a list lets you choose the right one instead of accepting the first match. It also shows you how the descriptions differ, which makes the next search more accurate.

Does it work for videos?

Yes. Uploaded videos are described and keyworded the same way as images, so a search can match a scene or a subject in a video rather than only a still. That is useful when your library holds product demos, behind-the-scenes clips and customer testimonials that were never named properly.

Video descriptions are naturally less precise than image descriptions, because a clip contains more than one moment. Describe the dominant subject or the setting rather than a detail that appears for a second. If you need to find a specific frame, the description is a starting point, not a timestamp search.

What are the limits?

A dark, blurry or very cluttered photo gives the model less to work with, and the description will be thinner. Very small objects in a wide shot can be missed. Text inside an image may or may not be picked up. None of this is a fault in the search; it is the reality of describing a picture automatically.

The practical habit is to check the Media page when a search fails, and to upload a clearer image when the description is poor. A sharp photo with one clear subject will always be easier to find than a crowded one. If the library is large, it is also worth removing duplicates so the same photo does not appear three times under three descriptions.

A filename is a label someone forgot to write. A description is what the picture actually shows.

Frequently asked questions

Do I have to tag photos myself?

No. Descriptions and keywords are generated when you upload the file. You can still add your own context elsewhere on the product, but the media search does not depend on you tagging every image by hand.

Can Ola find a video, not just a photo?

Yes. Uploaded videos are described and keyworded, so you can search by the subject or setting in the clip. A video description is less precise than an image description because a clip contains more than one moment.

Does the search use filenames?

The search is built on the content description, not on the filename. A file called IMG_4821.jpg can still be found by describing what is in it, which is the whole point of the feature.

Where do I see what the AI said about a file?

The Media page lists your files and their descriptions. Read the description there when a search misses, then use the words the description uses or upload a clearer image.

Which models describe the files?

Workers AI GLM 5.3 Flash is the primary model, with OpenRouter and Mistral Small 3.1 as fallbacks. The fallbacks keep descriptions flowing when the primary service is unavailable.

Content search turns a messy media library into something you can actually use. The habit to build is simple: upload a clear file, check the description on the Media page when a search misses, and describe the picture the way it looks rather than the way you named it. If you want to see the rest of what the assistant can do, read what Ola can actually do in your dashboard. For the platform, see features and pricing, or browse the Salesive blog.