Adobe has introduced AI Playground, a new experimental generative AI-powered editing experience in Adobe Project Indigo 1.1. The feature integrates AI photo editing directly into the experimental camera app, allowing users to modify images after capture without moving them to a separate editing application or web service.

Developed by Adobe’s Nextcam Team, Project Indigo is a computational photography camera app offering manual camera controls, JPEG and RAW capture, an SLR-inspired rendering approach, and computational photography features such as reflection removal when shooting through windows.

With AI Playground, Adobe is exploring how generative AI editing and photography assistance can become part of the camera workflow itself, allowing photographers to experiment with edits and receive guidance while they are still at the shooting location.

AI Playground is divided into four main sections: Object Editing, Styles, Photo Guidance, and Custom Edit.

Object Editing allows users to remove unwanted objects and distractions from images or blur backgrounds to create a shallow depth-of-field effect. Users can also specify custom categories of distractions using text or voice input and save them for future use.

Styles uses generative AI to apply different photographic and artistic looks to captured images, including styles inspired by traditional artistic techniques.

Photo Guidance analyzes captured images and provides suggestions for editing or reshooting them while users are still at the location. Adobe is also exploring device-specific recommendations based on individual smartphone cameras and lenses, including factors such as minimum focusing distance.

The company is additionally experimenting with composition and cropping suggestions using information captured through ultra-wide cameras or video.

Custom Edit provides a more open-ended prompt-based editing experience. Users can type or dictate their own instructions, edit prompts, and save frequently used commands. Repeating an operation generates another variation of the requested edit.

After taking a photo with Project Indigo, users can access AI Playground from the built-in filmstrip or full-screen gallery by tapping the sparkle icon. The feature currently works only with photos captured through Project Indigo.

AI Playground initially uses Google’s Nano Banana image-generation model for generative editing. Adobe said the underlying models could change over time, with different AI models potentially being selected for specific editing tasks.

Some capabilities may also use Adobe Firefly models, including tools designed to remove unwanted objects while preserving the remaining elements of an image.

The generative processing currently takes place in the cloud and requires an active Wi-Fi or cellular data connection. AI-generated images are returned at approximately 2K resolution to reduce processing and transmission times.

One technical challenge with the current implementation involves HDR photography. Project Indigo captures images in HDR, while existing generative AI models generally return SDR output.

To address this, Adobe has developed a custom machine learning model that converts the AI-generated SDR result back into HDR using the original HDR photograph as guidance. Adobe says the original image used during this process is not used for AI model training.

However, the company notes that HDR reconstruction may become less accurate when a generated edit significantly changes the colours or overall tones of the original photograph.

Adobe’s Nextcam Team is continuing to develop AI Playground with additional editing controls and photography guidance features. The company is also working on non-generative computational photography improvements, including better low-light image quality.

Another area under development is identity preservation. Current generative AI models can sometimes unintentionally alter facial characteristics or expressions when editing photographs. Adobe is exploring the use of reference photos and systems that could automatically identify suitable reference images to maintain a person’s appearance more consistently.

Adobe is also researching smaller on-device AI models that could eventually perform some generative tasks directly on smartphones. However, the company notes that current large image-generation models generally require cloud processing, while smaller distilled models offer more limited capabilities.

Adobe is working to add Content Authenticity Initiative (CAI) metadata to images edited through AI Playground. Some of the underlying generative AI models, including Nano Banana, may also apply invisible watermarks to generated content.

AI Playground is available as a free public experiment with Project Indigo 1.1, but access is currently limited to selected users who receive an invitation after updating and reopening the app.

No Adobe account or sign-in is required to participate, although users must opt in to anonymous usage analytics. Adobe says photos and prompts are not uploaded to Adobe’s own servers, reviewed by the company, or used to train its AI models.

Depending on the results of the experiment, Adobe may expand access, conduct additional tests, extend the feature, or eventually introduce a paid version.