IMatch AutoTagger: Using Hints and Context Placeholders

IMatch AutoTagger automatically adds descriptions, keywords, and custom traits to images using cloud or local AI. This post explains how to use hints, context placeholders, and existing metadata to guide the AI when processing large batches of images.

How AutoTagger Uses Prompts and Context

AutoTagger gives you several ways to guide the AI and work with its results:

• Use custom prompts to specify the format, length, level of detail, and style of the AI results.
• Include data from your IMatch database, such as location information or the names of people shown in an image, to provide context to the AI. This can improve descriptions and keywords.
• Control the keywords produced by AI to align them with your controlled vocabulary and thesaurus.
• Use the context placeholder to supply additional hints when running AutoTagger for a selection of files.

When to Store Hints in Image Descriptions

The context placeholder is useful when a set of files shares the same background information and you are working with AutoTagger interactively.

For images taken in Paris, for example, you can enter “This image was taken in Paris, France” as the context placeholder to help the AI generate more relevant results.

But what if a large batch contains groups of images that need different hints? Images may come from different folders or categories. The images in the folders 2026-05-03 (Paris) and 2025-08-03 (Rome), for example, need different location information.

One practical approach is to prepare hints before processing the batch. Store a hint in each file’s XMP description tag, then use a variable to include it in the prompt.

Example: Add Description Hints to an AutoTagger Prompt

Set the description for all images in the Paris folder to “This image was taken during our city trip to Paris, France, in May 2026”. For the Rome folder, use “This image was taken in Rome, Italy, in August 2025”. This takes only a few seconds in the Metadata Panel.

Include an IMatch variable in your prompt to pass the existing description to the AI:

{File.MD.description|hasvalue:Consider this existing description: "{File.MD.description}" when generating your description.}

If the image has a description, this variable returns an instruction containing it. Otherwise, it returns nothing. For the images in the Rome folder, it returns:

Consider this existing description: “This image was taken in Rome, Italy, in August 2025” when generating your description.

The prompt asks the AI to consider the existing description when generating a new one. This gives the AI context it might not be able to infer from the image alone and can improve the result.

The idea is to prepare hints for each group before letting AutoTagger process hundreds or thousands of images unattended.

What Happens When You Run AutoTagger Again?

If you often run AutoTagger again during review to improve a description, there is a catch: the original hint may no longer be there.
Your prompt uses the current description. If AutoTagger replaces your hint with the AI result, the next run feeds that result back into the prompt instead of the original hint. The AI gets its own homework back.

Use a Second Prompt

When running AutoTagger again, use a second prompt that does not include {File.MD.description}.

Use a Custom Tag or Attribute for Hints

Alternatively, store hints in another writable XMP tag or in an IMatch Attribute. Reference that tag or Attribute in your prompt instead of the description. As long as AutoTagger does not update the chosen field, the original hint remains available when you run it again during review. The XMP Instructions tag may be a good place for this, unless you already use it for other purposes.

Prepare Multiple Hints

A little preparation helps when running AutoTagger on large sets of images. Add the relevant hints to each group of files before processing them. You can also prepare separate hints for different tasks: one for the keyword prompt and another for the description prompt.

If the files have no keywords yet, you can store a hint as a keyword and include it in a prompt like this:

{File.MD.hierarchicalkeywords|hasvalue:Consider this description: "{File.MD.hierarchicalkeywords}" when creating keywords for this image.}

Use Existing Metadata as Hints

You may not need to write custom hints at all. Existing metadata in your IMatch database can provide useful context.

If you use IMatch reverse geocoding or enter location data manually, you can include that information directly in your prompt. For example:

This image was taken in {File.MD.city}, {File.MD.country}, in {File.DateTime|format:MMMM YYYY}

When the location and date fields contain the relevant values, these variables produce text such as “This image was taken in Rome, Italy, in August 2025” or “This image was taken in Paris, France, in May 2026”.

This provides an automatic hint for each file with the required metadata, giving the AI specific details to use in its descriptions. You can combine these variables with information supplied through the context placeholder or with hints prepared in advance.

For more examples, see the AutoTagger Prompting topic in the IMatch Help System.

Summary

Custom prompts let you guide the descriptions and keywords AutoTagger generates for your images.

Use the context placeholder when running AutoTagger interactively on a group of files that share the same context.

For large batches with different contexts, prepare hints for each group or use existing metadata. Reference those values through variables in your prompt so AutoTagger can process the batch unattended.