Adobe has introduced AI Playground, an experimental feature in Project Indigo 1.1 that adds generative AI-powered editing tools to its camera app. The feature allows users to edit photos after capture without transferring them to a separate editing application or web service.
Project Indigo 1.1.
Project Indigo was launched by Adobe’s Nextcam Team as an experimental computational photography camera app. It offers full manual controls, JPEG and RAW image capture, an SLR-like image rendering approach, and features such as removing reflections while photographing through windows.
Adobe said generative AI image editing has advanced, but prompt-based workflows still have limitations. Text prompts can produce different results depending on interpretation and context, often requiring multiple attempts. Existing workflows also require users to move images between camera and editing apps, making it difficult to edit photos while still at the shooting location. AI Playground integrates generative AI editing and photography guidance into the camera workflow.
AI Playground features
AI Playground includes four sections: Object Editing, Styles, Photo Guidance, and Custom Edit.
- Object Editing: Provides tools to remove unwanted objects and distractions, blur backgrounds to create a shallow depth-of-field effect, and remove selected distraction categories. Users can add custom categories through text or voice input and save or delete them.
- Styles: Applies photographic and artistic styles to captured images using generative AI models that can recreate traditional artistic techniques.
- Photo Guidance: Analyzes photos and provides editing and reshooting suggestions while users are still at the shooting location. Adobe is also working on customizing suggestions for individual smartphone models and lenses, including capabilities such as minimum focus distance. The company is exploring composition and cropping recommendations using additionally captured ultra-wide photos or videos.
- Custom Edit: Allows users to create edits using their own text prompts. Prompts can be typed or dictated, then edited, saved, or deleted. AI Playground buttons use Adobe-engineered prompts, and users can repeat an operation to generate another variation.
After capturing a photo with Project Indigo, users can open it through the built-in filmstrip or full-screen gallery and access AI Playground using the sparkle icon. The feature works only with photos captured using Project Indigo.
AI models and image processing
AI Playground initially uses Google’s Nano Banana image-generation model. Adobe said it may switch models over time or use different models for specific features. Some tools may also use Adobe’s Firefly models, including object removal capabilities for removing distractions while preserving other parts of an image.
The feature uses cloud-based AI processing and requires an internet connection through Wi-Fi or cellular data. Generated images are returned at approximately 2K resolution to reduce transmission and processing time.
Additional technical details include:
- If generation fails due to server issues, users can retry the request.
- Images or prompts that violate Nano Banana’s AI safety policies may be rejected.
- Repeated safety violations may result in removal from the experiment.
- Project Indigo captures HDR photos, while current generative AI models output SDR images.
- Adobe uses a custom machine learning model to convert SDR AI-generated output back into HDR using the original HDR photo as guidance.
- The original HDR image is used only for HDR enhancement and is not used for model training.
- HDR reconstruction may be less accurate when AI-generated images have significantly different colours or tones from the original image.
Future development
Adobe is continuing to develop AI Playground with additional editing tools and photography guidance features focused on common editing tasks and prompts. The Nextcam Team is also working on non-generative AI computational photography improvements, including low-light image quality improvements.
Areas under development include:
- Generative editing controls: Adobe is exploring additional controls and workflows as current prompt-based systems require users to modify prompts and generate multiple versions.
- Identity preservation: Generative models can alter facial features or expressions in edited images. Adobe is exploring reference-photo-based approaches and technologies that can automatically identify suitable reference images.
- On-device AI models: Adobe is researching smaller AI models that can run directly on smartphones. The company said current large image-generation models are generally cloud-based due to their size, while smaller models created through distillation currently provide fewer capabilities.
Content authenticity and availability
Adobe is working to add Content Authenticity Initiative (CAI) metadata to AI-edited images created through AI Playground. Some AI models used by the feature, including Nano Banana, also apply invisible watermarks to generated images.
Project Indigo’s AI Playground was developed by Mohammad Haque, Boris Ajdin, Ruiming Cao, and Marc Levoy with support from the Nextcam Team.
Key availability details include:
- Release: Available as a free public experiment in Project Indigo 1.1.
- Access: Limited to selected users who receive an invitation after updating and reopening the app.
- Account requirement: No account or sign-in is required.
- Analytics: Participants must opt in to share anonymous usage analytics, including button presses.
- Data handling: Adobe said photos and prompts are not uploaded to its servers, reviewed by the company, or used to train AI models.
- Future rollout: Depending on the results, Adobe may extend the experiment, expand access, conduct additional tests, or introduce a paid version.