Search for assets
With AI Visual Search, finding the perfect asset is as easy as talking to a colleague.
Simply ask for "pictures of a beach", and AI Visual Search will return all matching assets in your library without requiring you to remember any of the associated keywords and tags. Canto's exclusive AI technology can even interpret more complicated and abstract requests, such as "show me a happy child" or "a short clip of a busy intersection", matching them against the detailed catalog of visual data for each individual asset.
|
Please note: AI Visual Search can only include already indexed assets in a search. Once added to your account, Canto's AI will automatically perform an initial mapping of your Main Library. This process may take one to two weeks, depending on volume of assets in your storage, and does not include content in the Trash Bin. Restoring deleted assets that were not captured during this time will not automatically index them; instead, they will need to be re-uploaded. AI Visual Search will automatically index any new uploads. Depending on the number and size of assets to be indexed, this process may require additional time to complete. |
Best practice for visual searches
Classic search engines often require highly simplified search terms to ensure that they can provide accurate results. With the rise of AI, this way of searching for content may cause search algorithms to perform at a lower level than intended.
AI Visual Search is trained on natural language, which means that it not only understands when you address it as if you were speaking to a human colleague, but it expects this type of prompt in order to provide the best performance. To ensure that you are using AI Visual Search at its full potential, make use of the following list of categories as you formulate a prompt. The more details you provide in your request, the more accurate your results will be.
- Subjects and objects in the scene
- Actions performed
- Time of day
- Emotion
- Lighting
- Composition
- Colors
- Perspective
It is important to remember that AI Visual Search does not possess any biases when analyzing an image and will take note of details that a human may not immediately consider. While a user might not notice that the image of a crowd happens to include a dog, AI Visual Search will still associate the asset with the term "dog" in the same way that a portrait showing only a dog would.
Which asset is prioritized when searching for the term "dog" depends entirely on the 'weight' of the keyword, as determined by the context of the prompt and other elements in the image. If the provided request is too simple, AI Visual Search will lack the details to be able to assign an appropriate weight and the images will likely be returned in a random order. Once more detail is added – for example, by specifying "dog in a crowd" – AI Visual Search can determine the appropriate weight and prioritize the matching asset in its results.
The following examples show possible search terms, and their breakdown based on these categories.
|
Search term used: "planes with contrails and a blue sky background" Breakdown: Subjects and objects in the scene - planes, contrails, sky Composition - planes with contrails, sky background Colors: blue sky Search term used: "women's blue summer sleeveless pajamas" Breakdown: Subjects and objects in the scene - pajamas, and the qualifying variations sleeveless pajamas, summer pajamas, women's pajamas Colors: blue Search term used: "Person adding colored sticky notes to a whiteboard" Breakdown: Subjects and objects in the scene: person, sticky notes, whiteboard Actions performed: adding colored sticky notes Colors: colored sticky notes, whiteboard Search term used: "close up of a happy family laughing together" Breakdown: Subjects and objects in the scene: family Actions performed: laughing Emotion: happy Composition: close up, laughing together |
Enhance results with metadata
Where other search solutions are limited to either a standard metadata search, or a search based on visual data, AI Visual Search combines both into a single function and compares your prompts not only against its own image data, but also against the metadata of your assets, albums, and folders.
It then determines the relevance of an asset based on the best match to your search input.
|
Example: User A is looking for assets related to make-up. They have entered the prompt "make-up palette", which returned a variety of results. AI Visual Search sorted these to display the album "Make-up Palette" in the first position, as its name is an exact metadata match. Click to zoom
|
You can optimize your search results by adding specific metadata – such as tags, SKU numbers, or file names – to a natural language prompt.
|
Example: User A is searching for an asset from a specific make-up set product line. They first enter the prompt "make-up palette set on a white background." While this generates a large number of visual matches, the specific asset is not among the top results. Click to zoom
User A then remembers that the product has the SKU number "9012" listed in the metadata and amends their prompt to "make-up palette set on a white background 9012." AI Visual Search includes this in its analysis and prioritizes the assets with matching metadata to appear in the top results. Click to zoom
Click to zoom
|
When adding multiple metadata points to a prompt, you can use quotation marks to indicate which of them is most important to you. AI Visual Search will then prioritize any results that contain an exact match to your term in any of their metadata fields.
|
Example: User A is now searching for an asset of a make-up palette from a specific product line that must contain a specific color. Knowing that their tenant is configured to automatically recognize colors in images, they include the HEX value of the specific color in their prompt, searching for 'make-up palette set on a white background UPC9012 "#EC3292"'. By placing the HEX value in quotation marks, AI Visual Search prioritizes the color as the most crucial part of the prompt and lists assets containing this value in their metadata among the top results. Click to zoom
Click to zoom
|
Find video clips
AI Visual Search not only catalogs and searches static image assets, but also analyzes video frames to find clips that match your prompt. You can emphasize the type of file AI Visual Search should prioritize by including phrases such as "a clip showing" or "video of" in your search. Depending on the length and content of a video, AI Visual Search may find multiple matching clips in a single asset.
Identifying video assets
In the gallery, video assets can be identified by the -symbol, and will display the
-symbol followed by a number in their top left corner, indicating how many matching clips were found in the specific asset.
Edit and download clips
When you click a video asset found through AI Visual Search, it will open a modified version of the Preview Page.
The video preview will contain an additional section, which lists the found clips. Each clip will be represented by a thumbnail of its initial frame, along with the clip's start time and duration. AI Visual Search automatically sorts clips by the quality of their match, and the clip with the highest priority will be marked as " Best Match" ①. Clicking a thumbnail will skip the video playback to the start position of the clip – marked with the
-symbol on the timeline ②. Click the
-button in the top left ③ to open the "Download File" overlay.
To trim your chosen clip, select "Edit Clip" on the right ④.
To change the start time, move the -playhead along the timeline or use the
-buttons. The end point of the clip can be adjusted using the
-playhead. Moving the
-playhead below the timeline will update the preview to show the frame at that position. ⑤
Click "Download" ⑥ in the bottom left to save your changes and download the clip.
Sort and filter
As your library expands, combing through search results to find an asset that not only matches your prompt but also meets other specifications required for your project can be a time-consuming task. To help speed this process up, both the basic Keyword Search and AI Visual Search provide various ways for further refining search results.
The Filter panel to the right of the gallery allows you to remove results that do not match a specific selection of metadata, such as file-type, image dimensions, approval status, or upload date. To learn more about filters, please visit the article "Apply Filters".
You can also sort results using the menu located below the search bar. In addition to the default sorting options – which you can read more about in the article "How to: Sort content in the gallery" – AI Visual Search provides the following two alternative methods.
AI Relevance
This sorting method ranks the results based on the level of certainty that AI Visual Search has regarding the match between the visual content of the asset and your prompt. This method exclusively sorts from most relevant to least relevant asset.
|
Example: User A has entered the prompt "a couple holding hands while hiking in a forest". AI Visual Search has returned a large number of results; however, most are only partial matches or do not seem to be matches at all. As User A looks over the assets, they notice that the results are sorted by name. Click to zoom
To improve the returned results, User A changes the sorting method to "AI Relevance". After the gallery refreshes, the top result is an exact match for the phrase "a couple holding hands while hiking in a forest," whereas assets that are listed lower in the results have less overlap with the prompt, such as showing a couple in a forest but not holding hands. Click to zoom
|
Clip Count
This sorting method ranks the results according to the number of matching clips contained within each asset. It can either sort from high to low, or low to high.
|
Example: User A is looking for clips of "Lego bricks being assembled". AI Visual Search has returned a large number of results and sorted them by relevancy. However, despite being good matches, most of the assets are static images. Click to zoom
User A changes the sorting method to "Clip Count", allowing them to easily find the matching video assets, and see which one will have the most content available. Click to zoom
|
Switch search methods
To ensure that you always have the best tools available, Canto lets you seamlessly switch between AI Visual Search and the traditional Keyword Search method. This allows you to choose the approach that aligns best with your preferences and requirements at any given moment.
|
Please note: AI Visual Search is a separate add-on feature. If you are interested in gaining access to this extremely powerful tool, do not hesitate to reach out to your Account Manager, or contact our customer service team. |
Check if AI Visual Search is enabled
You can verify the current search method by checking the status indicator located on the right side of the search bar, above the gallery.
"AI Enabled " indicates your current search method is set to AI Visual Search ⑦. Any entered prompt will be analyzed using natural language recognition, and then matched against the catalog of visual data for both static images and video clips, as well as any applicable metadata.
"AI Disabled " indicates your current search method is set to Keyword Search ⑧. Searches will be limited to available asset metadata, and only results with exact matches to the prompt will be returned.
Toggle AI Visual Search on or off
Switching between AI Visual Search and the Keyword Search is easily done from anywhere within the Main Library.
First, verify which search method is currently active by checking the status indicator in the search bar above the gallery. Then, simply click the indicator to turn AI Visual Search on or off.
|
Please note: Changes made to the active search method on an individual user account are non-permanent and will revert to the default search method, as defined in the tenant's settings, whenever the active browser tab is refreshed, the user navigates away from the library, or the user logs out of their account. |
Set your tenant's default search method
|
Please note: The actions described in this section require Administrator privileges, and are unavailable on Consumer and Contributor profiles. |
You can adjust your tenant's default settings to ensure that the search method best suited to your team's needs and preferences is always immediately accessible
Open the "Profile & Settings" menu by clicking your profile icon – by default this will be the symbol, unless a custom profile picture has been set. Then, select "Settings".
Select the tab "Configuration Options", and choose "Defaults" on the left.
In the "Defaults" section, select the tab "Search & Filter Panel". Under "Global Search Type" select either "AI Visual Search" or "Keyword Search". Click "Save" in the top right corner to confirm your choice.