Adobe’s ‘natural look’ camera app embraces generative AI

Adobe’s Indigo Camera App Leans Into Generative AI — and Skips Its Own Models

Adobe’s experimental Project Indigo camera app is pivoting hard into generative AI. Originally launched last year as a tool for iPhone photographers who wanted a “more natural (SLR-like) look,” the app’s version 1.1 update introduces a suite called AI Playground built on Google’s Nano Banana AI model — not Adobe’s own Firefly. For developers watching the intersection of mobile photography and generative tools, this is a revealing case study in how even deeply AI-native companies choose their infrastructure, design their user experience, and hedge their bets on what “AI photography” actually means.

The AI Playground includes four features: Styles, which applies AI-generated filter effects (illustration looks, lighting changes); Object Editing, which removes people or objects and fills the background with generative AI (similar to Google’s Magic Eraser); Photo Guidance, which analyzes photos to suggest reshoots or edits; and Custom Edit, a broader prompting tool that lets users describe specific changes. Crucially, Adobe describes each feature as containing “multiple buttons or toggles, which trigger GenAI with prompts we have engineered for them.” If a user dislikes a result, pressing the button again “typically get[s] something slightly different.”

Adobe is testing these tools with a small percentage of Indigo users over the next few weeks, offering free access with no sign-on required. There is an opt-out button for those who want to avoid AI editing altogether. Adobe Fellow Marc Levoy stated the company wants to “learn how people use GenAI-powered editing,” and if the feature proves popular, Adobe will eventually offer a paid version.

Why Google’s Nano Banana Instead of Firefly?

The most striking decision is the model choice: Google’s Nano Banana AI, not Adobe Firefly. Adobe says it is using that model “to start” and may switch or add others later. For developers, this signals that model selection is not purely a matter of brand loyalty or vertical integration. Nano Banana is likely optimized for on-device or low-latency inference, which matters enormously for a camera app where users expect near-instant results after pressing a button. Firefly, while powerful, may be too heavy for mobile use cases without significant cloud round-trips. The result is a pragmatic decision: use whatever model delivers the best performance for the specific interaction patterns Adobe wants to test.

This has a second implication: vendor lock-in is not inevitable. By making the AI backend swappable, Adobe future-proofs the app’s architecture. Developers building similar tools should consider abstraction layers that let them evaluate models independently — the best model for object removal may differ from the best model for style transfer, and both may change in six months. Adobe’s approach suggests that modularity matters more than betting on a single provider.

UX Lessons for Generative Features

The AI Playground design has several notable UX decisions worth dissecting. First, the use of pre-engineered prompts behind simple buttons (“Press for a style, press again for a new variation”) collapses the complexity of prompt engineering into a single tap. This is a smart pattern for consumer photography apps: it removes the friction of writing prompts while retaining the generative variety that makes AI tools feel responsive. Users get the exploration of a random roll without the cognitive load of a text box. Developers can replicate this by creating curated “prompt vectors” that map to specific visual effects, then randomizing or varying parameters (like seed or guidance scale) on subsequent taps.

Second, the opt-out mechanism is a safety valve. Adobe explicitly frames AI Playground as an experiment — users who dislike it can continue using the app as before. That reduces the risk of backlash from the “real photography” community (already a sensitive topic for Adobe, as the article notes with the line “Welcome to the ‘what is a photo’ debate”). For any developer adding generative AI to a product that previously offered a traditional experience, having a clean off-ramp is critical. It also provides a control group to measure whether AI features actually drive engagement or cause abandonment.

Third, the Photo Guidance feature — where AI analyzes photos and recommends reshoots or edits — is a particularly interesting direction. It moves AI from an editing tool to a creative coach. That kind of feedback loop could change how users learn photography, reducing reliance on tutorials and tutorials. Developers building camera or photo apps should consider whether their AI can suggest actions rather than just apply them. It’s a small step from “try shooting from a lower angle” to “I will adjust this for you,” but the former keeps the user in control.

How This Fits the Wider AI Imaging Landscape

Adobe’s move echoes trends across the industry. Google’s Magic Eraser was one of the first mainstream implementations of generative fill for mobile photography. Apple has been adding machine learning-powered editing to its Photos app for years. But Indigo’s AI Playground is notable because it layers multiple generative capabilities — filters, object removal, coaching, custom prompts — into a single experimental suite rather than rolling them out one by one. That indicates Adobe is treating this as a cohesive product experiment, not just a feature test.

For developers choosing between AI image generation tools, the field is only getting more crowded. Adobe’s reliance on a third-party model here (even temporarily) underlines that the best tool for a given job depends heavily on latency, device support, and model size. To stay current on the full range of options, developers can check our detailed comparison of the best AI image generators available in 2026. The trade-offs between cloud-based models like Firefly and on-device models like Nano Banana are central to that decision.

Monetization and the Paid Version Question

Levoy’s statement about a potential paid version reveals Adobe’s typical freemium playbook. Free access with no sign-on is deliberately low-friction, designed to maximize data collection on how users interact with each tool. The goal is to identify which features have enough perceived value to convert users to paying customers. For developers, this is a classic MVP pattern: launch a limited, free experiment; measure feature usage and retention; then build a monetized product around the most successful pieces. The key insight is that the AI Playground is already a paid feature waiting to happen — model inference costs money, so Adobe will eventually need to recoup that. But by testing with a small pool first, they reduce infrastructure costs and learn the demand curve before scaling.

Developers building similar features should consider metering usage carefully. If you give away unlimited AI edits, you eat inference costs and devalue the feature. If you charge upfront, you may deter the very users whose behavior you want to study. Adobe’s approach — free but temporary for a limited audience — is a reasonable middle ground, though it risks creating backlash if users become dependent on the free tool before it gets paywalled.

What Developers Should Watch Next

Adobe’s Indigo experiment is still in its earliest phase, but several signals matter for the broader developer community:

• Model flexibility: Adobe explicitly says it may switch models. Build your stack to be model-agnostic from the start; you do not want to rewrite your inference layer every quarter.
• Button-based prompts: The “press for variety” pattern works well for novice users and could be adapted for any generative tool — text, image, or audio.
• AI as a coach vs. AI as a tool: Photo Guidance suggests a shift from “edit my photo” to “tell me how to shoot better.” This may unlock new UX paradigms for educational apps.
• Opt-out is a feature, not a concession: Giving users an escape hatch from AI features builds trust and provides clean telemetry on feature adoption versus rejection.

Meanwhile, the conversation around AI chatbots continues to evolve alongside these visual tools. For teams building conversational AI interfaces into their products, our ranking of the best AI chatbot platforms in 2026 offers a parallel perspective on model choice, deployment, and user experience design.

Ultimately, Project Indigo’s AI Playground is not just about a camera app — it’s about how a company with deep AI resources (Firefly) decides to use a different model for a specific product, how it structures an experiment to validate demand before committing, and how it designs generative interactions to feel predictable despite inherent randomness. For developers building any kind of AI-forward application, these decisions are the actual product. Copy the patterns, not just the features.

Source: The Verge. Details as reported; verify specifics at the source.