The New Adobe App Critiques Your Photos and Tells You How to Reshoot Them on the Spot

A year after launching Project Indigo, the iPhone photography app aimed at delivering a "natural" look and feel, Adobe has added tools that move in the opposite direction: a suite of generative AI. The Nextcam team calls it AI Playground, and it is an experiment within an already explicitly experimental app.

The goal is to "explore how generative AI can assist photographers at the moment of the shot," meaning while still in front of the scene. For now, the feature is free and does not require registration, but it will only be available to "a few percentage points" of users and for "a few weeks"; participants agree to share their interactions with the buttons, while Adobe assures that it does not view or upload prompts and images, nor uses them for model training. If the feature proves successful, it will then become a paid service.

Four Sections, Almost All Button-Based

AI Playground appears with version 1.1 and works only on photos taken within Indigo: the star-shaped button remains grey on others. It is divided into four sections.

Object Editing removes distractions, with switches for pre-set categories (background people, cables and poles, vehicles, fences, garbage) or described by voice, and simulates a reduced depth of field.
Styles reimaginates the shot by imitating artistic techniques, from watercolor to ink drawing.
The most interesting section is Photo Guidance: a language model provides a critique of the just-taken photo (framing, lighting, colors, emotional impact) and suggests how to reshoot or modify it. This is a function that, as Adobe notes, makes sense only within a camera app, because advice comes while still on-site. The natural comparison is with Google's Pixel Camera Coach, but here the judgment is more descriptive and comes afterward.
The fourth section, Custom Edit, allows for free prompts, which can also be dictated by voice. It even allows for the reverse process concerning styles: from a photo of a painting or sculpture, the app tries to derive a photorealistic image. Marc Levoy, an Adobe Fellow who previously enhanced the photographic capabilities of Google’s Pixels, recounts having elicited "astonished looks" when showing the photographic version of a bronze sculpture while it was right in front of him.

The Model is from Google, Not Firefly

AI Playground runs on Nano Banana, a Google model based on Gemini, and not on Adobe's Firefly, which could potentially be integrated for individual buttons, being particularly adept at removing distractions without affecting the rest. It is a cloud-based model, so a connection is necessary, and the returned images are about 2,000 pixels per side. Since Nano Banana only produces standard dynamic range images, Adobe has added its own model to bring them back to HDR using the original shot as a guide, with results that it claims are not always accurate.

There remain the known limitations of generative editing, which Adobe lists bluntly: no slider to control the outcome except by rewriting the prompt, the tendency of models to alter unrequested details such as the shape and expression of faces, and the slowness of large cloud-bound models. On the transparency front, the team says it is working on adding Content Credentials C2PA, while Nano Banana already applies Google’s invisible watermark SynthID; for the rest, Levoy places the responsibility on the user: whether a modification is "correct or misleading," he writes, depends on how it is used and presented.