Adobe is adding AI-powered feedback tools to Project Indigo, its experimental iOS camera app. The new features use large language models to critique photos and suggest ways to improve them, both before and after you shoot. According to TechCrunch, the update also brings advanced object removal, depth of field simulation, and style transfer options.
Project Indigo launched last year as a testing ground for Adobe’s camera ideas. It already had pro controls, multi-frame super-resolution, and various capture modes. This update pushes it further into AI territory, with tools that go beyond the basic auto-enhance buttons found in most photo apps.
Marc Levoy, who leads the project and previously worked on the camera system inside Google’s Pixel phones, has a specific view on how AI editing should work. He thinks prompt-based editing is frustrating because finding the right wording to get the result you want is harder than it sounds. Most of the new features are buttons that produce predictable, consistent outputs rather than asking you to type instructions and hope for the best.
The two standout additions are built around giving users actual feedback on their photos. The first is a photo critique tool that offers a “professional” opinion covering framing, lighting, colors, and emotional impact. The second is a capture and edit suggestions tool that tells you how to reshoot a scene, whether that means adjusting framing, changing exposure, or moving objects out of the viewfinder. In one example from testing, the app flagged a specific hexagonal white object in the frame and suggested removing it.
There is also a second layer to the edit suggestions feature. After analyzing an existing photo, the app tells you how to improve it using Adobe Lightroom controls, which could help users actually learn what tools to reach for rather than just applying a filter and moving on.
This matters because most AI photo tools treat editing as a black box. You press a button and something happens. Project Indigo’s approach is more instructional, which is a meaningful shift if you care about understanding photography rather than just automating it.
The object removal feature is also worth noting. Apps like Apple Photos, Google Photos, and Adobe Photoshop have offered this for years, but they typically require you to manually circle or draw around the object you want to remove, which does not always work cleanly. Project Indigo handles this differently. It gives you a set of toggles for common objects to remove:
- People in the background
- Trash and trash cans
- Wires and poles
- Fences
- Vehicles
- General clutter
You can also describe a custom object if it does not fit one of the presets. In testing, the app removed a person and the object they were holding from a photo background without leaving visible artifacts, which is a stronger result than many existing tools manage.
The depth of field feature lets you use AI to simulate a blurred background, and the style transfer option can apply looks like watercolor, pen and ink, monochromatic, and backlit subject. The style transfer results look polished, though the feature itself is not new. Apps like Prisma were doing this years ago, and it has limited practical value for most photographers.
Project Indigo also includes a freeform text prompt feature for edits that do not fit the preset tools. Levoy has been critical of prompt-based editing, so its inclusion feels contradictory, though Adobe appears to be offering it as a fallback rather than the main approach.
On the technical side, Adobe is currently using Google’s Gemini Nano Banana model to power these features, but the company says it is open to switching to other models, including its own Firefly model. All of these tools sit inside an “AI playground” tab and are only available to a limited group of testers for now. There is no guarantee they will ever reach a wider audience.
That said, the direction is interesting. A lot of AI photo features are focused on making images look better without teaching you anything. Project Indigo is at least trying to make users more aware of what makes a photo work, which is a more useful goal even if the execution is still experimental.
