Adobe Camera App's New Feature Will Critique Your Photos Using AI

Adobe dropped a significant update to its experimental iOS camera app Project Indigo on July 20, 2026, introducing an AI-powered photo critique feature that gives users structured feedback on framing, lighting, color, and emotional impact — all without asking them to type a single prompt. The update, which Adobe is calling "AI Playground," also brings category-based object removal and depth-of-field generation, and it runs on Google's Gemini Nano Banana model rather than Adobe's own Firefly AI.

What's Converging

The photography software industry has spent the last two years cramming generative AI into every available slot. Apple Photos added AI cleanup tools, Google Photos added Magic Eraser improvements and AI-assisted search, and Adobe itself shipped Firefly-powered generative fill in Photoshop. What's notable is that none of these integrations have seriously tried to teach users anything — they just execute. The assumption has been that users want results, not coaching.

That assumption is starting to crack. Google launched a camera coaching feature for Pixel phones last year, offering framing suggestions in real time. But multiple reports this week point out that Google's coaching stayed at a fairly high level of generality — useful for broad feedback, less useful for learning the craft. Adobe's approach, by contrast, breaks critique into named dimensions and then extends the loop by connecting those observations directly to editing actions, either via reshoot suggestions or specific Lightroom control adjustments. That's a more complete feedback cycle than anything the competition has shipped publicly to date.

The broader pull here is that AI tools are gradually moving from "do the thing for me" toward "teach me to do the thing better." Developers building feedback systems — in design tools, writing assistants, and code reviewers — are discovering that unstructured AI output tends to be long on encouragement and short on specificity. Structured evaluation schemas, where the AI is constrained to score or comment on a fixed set of named dimensions, tend to produce more actionable and consistent results. Adobe's choice of architecture here isn't just a UX preference; it reflects a design principle that's worth watching across the industry.

The Specific Development

Project Indigo launched last year as a deliberate counterpoint to the overprocessed look of most iPhone camera apps. It offered full manual controls, multi-frame super-resolution, and capture modes aimed at producing images with a more SLR-like quality. The new AI Playground update builds on top of that foundation without replacing it — Adobe included an opt-out button that lets users keep using the app as before, and The Verge reports that free access to the AI suite is being tested with no sign-on requirement, reaching a small percentage of Indigo users over several weeks.

Marc Levoy, who led camera development on the Google Pixel and is now heading Project Indigo, made an explicit design choice to move away from prompt-based editing. The reasoning, as reported by TechCrunch, is that getting the right result from a freeform AI prompt is itself a skill, and one that most users don't have. By replacing prompts with buttons that generate more consistent outputs, Adobe trades flexibility for reproducibility. The photo critique feature reflects this philosophy: rather than asking users to query the AI about their image, it automatically structures feedback into four named axes — framing, lighting, colors, and emotional impact — and then surfaces specific Lightroom controls for acting on those observations. A second tool handles real-time capture suggestions, pointing out what to change in the viewfinder before the shot is taken.

Object removal has a similar treatment. Rather than requiring users to draw around the object they want removed — a process that works inconsistently depending on the selection quality — Indigo's version lets users pick from a category list: people, trash, wires, vehicles, and similar common distractors. Digital Trends and TechCrunch both note that the features run on Google's cloud-based Nano Banana model, which requires an active internet connection, and that edited photos are capped at 2K pixels to keep response times reasonable. Adobe has flagged that repeatedly triggering the model can produce inconsistent results, and the company has been transparent that this is experimental territory. Firefly integration is planned for a future phase, making the current Nano Banana dependency a starting point rather than a commitment.

Our read is that the 2K output cap is the most practically limiting detail here, and one that matters more than the coverage has acknowledged. Photographers who use Indigo precisely because it produces high-quality captures are now being asked to route AI-edited versions through a ceiling that sits well below the native resolution of modern iPhone sensors. That trade-off is probably acceptable for learning and experimentation, but it draws a clear line between Indigo's AI Playground and a professional-grade editing workflow.

What's Likely Next

The immediate question is whether AI Playground survives the test window and reaches all Indigo users. Adobe's language — "depending on its success" — makes the broader rollout explicitly conditional. The company is running this as genuine user research, not a staged feature launch, which means the critique dimensions, the category toggles for object removal, and even the Nano Banana model dependency could all change before anything goes wide. Developers building similar feedback tools should watch the rollout for signals about which dimensions users actually engage with versus which they skip.

The longer arc involves the Firefly integration Adobe hinted at. Nano Banana is a capable multimodal model, but Firefly carries Adobe's own data governance and creator-rights framework — a meaningful distinction for a company whose commercial user base is professionally sensitive about training data provenance. If Adobe can route the structured critique schema through Firefly rather than a third-party cloud model, it closes a gap that's currently visible in the fine print. Whether that swap preserves the quality and speed users see from Nano Banana is the engineering question that will determine how seriously the photography community takes AI Playground past its experimental phase.

Sources