What the Assistant Can and Can't Do (Yet)

An assistant that can “find any photo instantly” sounds like it can do almost anything, and marketing copy for AI features tends to lean into that impression. We would rather tell you exactly what Imagen’s assistant does today, what it does not do yet, and where the honest edges are, so you know what to expect the first time you try it instead of guessing.

What the assistant is built to do

The assistant lets you describe what you are looking for in plain language and get results without digging through a gallery yourself. Ask for every frame rated five stars, or describe a moment you remember, like a particular exchange during a toast, and it searches through the shoot for you. It is connected to the tools already inside Imagen, so once it surfaces a set of photos, you can act on that set directly rather than copying results somewhere else first.

That combination, plain-language search plus a direct connection to your other tools, is the actual point of the feature. It is not meant to replace your review, only to get you to the relevant photos faster so the review itself takes less time.

What it knows: content, not location

Here is the most important limit to understand. The assistant knows what is in a photo, not where within the photo that thing is. It can tell you a frame contains a smiling couple, but it cannot yet point to the exact pixels that make up a face or a sky in order to place a mask there for you.

In practice, that means the assistant is excellent at helping you find and filter, and not yet able to do spot-specific editing on your behalf. If you need a mask on a particular area, that is still a manual step in the editor for now, separate from anything the assistant does.

It does not generate images

This is worth stating plainly: the assistant does not create or invent photos. It works only with the real images you actually captured and uploaded. Ask it to find something that never happened in your shoot, and it will not manufacture a result, because there is nothing for it to find. Every photo it surfaces is one you took yourself, nothing more and nothing less.

No voice input, at least for now

You type your requests rather than speak them. There is no voice input in this version, so every search or instruction goes through text. For photographers used to working with a keyboard already open, this tends to be a small adjustment rather than a real obstacle, but it is worth knowing upfront if you were picturing a hands-free experience during a busy edit session.

The Imagen assistant finding and editing photos from a plain-language request

Subjective requests can be a stretch

The assistant does well with concrete, describable requests: ratings, specific actions, visible details. It can struggle with anything highly subjective, like finding “the most emotional photo” or “the best shot of the dance floor,” because those judgments depend on taste in a way a written description cannot fully capture. If a request comes back thin or slightly off, try making it more concrete rather than more clever, and the results generally improve.

A closer look: phrasing a request well

Say you shot a five-hundred-person conference and need every photo of a specific keynote speaker for a client’s recap deck. Asking the assistant to find that speaker by a visible detail, like what they were wearing or roughly when they were on stage, tends to work well, because that is a concrete, describable request. Asking it to find “the best shots” of the whole event is a much harder ask, because “best” is a judgment call, not a visible detail. The difference between those two requests is a good preview of where the assistant’s current strengths and limits actually sit, and it is a useful habit to carry into any request you make.

Why we are telling you this upfront

It would be easy to describe the assistant only in terms of what it can do and let you discover the limits on your own. We think that wastes your time and sets the wrong expectation. Knowing that it works on content rather than location, that it does not generate images, and that very subjective asks are harder, means you can use it well from the first session instead of testing its edges by trial and error.

None of these limits are permanent. The assistant is one of the areas expected to keep growing, and capability here should expand over time. We are not attaching dates to any of that, because we would rather ship improvements when they are ready than promise a timeline and miss it.

Getting good results today

Be specific about what you want: a rating, a visible action, a rough part of the shoot, a described subject. Combine the assistant with smart tags and manual ratings rather than treating it as the only tool for the job, since each one covers a slightly different part of finding the right photo. A little practice with phrasing goes a long way, and most photographers find their results improve noticeably after just a handful of tries.

Remember that it is connected to the rest of your project, so once it finds a set of photos, the next step, whether that is exporting, rating, or editing, is right there rather than somewhere else entirely. For more context on how the pieces of the new Studio fit together, our valuable tips library is a good next stop.

It does not know your business context

The assistant works from what it can see in a photo and from what you tell it directly. It does not know your contract terms, your client’s specific preferences from a call you had last week, or which shots a venue asked you not to publish. If a request depends on information outside the photo itself, plan to supply that context by hand, either by building it into how you phrase a request or by applying it yourself after the assistant returns a set of photos.

This is a natural limit rather than an oversight. A tool that searches your photos does not need to also manage your client relationships, and keeping those concerns separate makes the assistant more predictable, not less capable. Think of it as a search tool with excellent recall and no memory of your business beyond the current shoot. That trade-off is deliberate, and it is what keeps the assistant fast and consistent across every project rather than trying to be a general-purpose business tool.

A second closer look: when a first request falls short

Sometimes the first phrasing of a request comes back thin, missing a few frames you know exist, or including some that clearly do not fit. Before assuming the assistant cannot find what you need, try narrowing the description: add a visible detail you left out, mention roughly where in the shoot the moment happened, or split a broad request into two smaller ones. In practice, most disappointing results improve significantly on the second try once the request gets more specific rather than more general. The same principle applies to almost any request you might think to make, not just a search built around a single visible detail.

Where this is heading

The gap between what the assistant can do today and what photographers eventually ask for is exactly where future updates are aimed. Location-aware requests, more nuanced subjective judgment, and broader voice support are all reasonable directions, though none of them are promised for a specific release. For now, the honest description is the useful one: a fast, capable search layer for concrete requests, with real limits around subjective judgment, spot editing, and generated content.

None of that changes what is true today, which is the point of this whole page: use the assistant confidently within its current strengths, and treat anything beyond that as a bonus rather than an expectation you were promised.

What’s new in Imagen: The assistant is a good place to start with the new Imagen, an intelligent, agentic Studio for professional photographers where you cull, edit, review and export in any order. It finds photos and acts on them through Imagen’s tools when you ask in plain language. It knows what is in a photo but not where, it does not create images, it has no voice input, and very subjective requests can be a stretch. The Studio is rolling out gradually.

Frequently asked questions

Can the assistant place a mask on a specific part of a photo?

Not yet. It knows what is in a photo but not where that thing is located within the frame, so spot-specific masking is still a manual step in the editor.

Can I create new images by asking the assistant?

No. The assistant only works with the real photos you have already captured and uploaded. It does not generate or invent imagery.

Can I talk to the assistant instead of typing?

Not currently. Requests go through text, and there is no voice input in this version.

Why does the assistant sometimes miss what I meant?

Very subjective requests, like asking for the “best” or “most emotional” photo, are harder for it to judge than concrete requests like a rating or a specific visible detail. Making a request more specific usually helps.

Curious what the assistant can find in your own galleries? Try it for yourself in Imagen’s new Studio, where you cull, edit and export in one project, at imagen-ai.com.