Adobe AI Assistant for Photoshop: What It Can Actually Edit
PhotographyThe gap between describing an edit and executing it in Photoshop has always required fluency in the tool itself. Adobe’s new AI Assistant changes that boundary: users describe what they want in plain language, and the assistant either executes it automatically or walks through the steps. The public beta is live now on Photoshop web, iOS, and Android — though not yet on desktop.
What It Can Do
The capability list covers the edits photographers actually request most: removing unwanted people or objects, changing backgrounds, refining portraits, adjusting lighting and color, and cropping for specific aspect ratios or platforms. These aren’t text-to-image generation tasks — they’re edit operations on existing photos, executed through the same Photoshop tools that previously required menu navigation and manual adjustments.
Two interaction modes are available. The first is automatic execution: describe the edit, the assistant runs it, you review and accept or reject. The second is instructional: the assistant explains what to do step by step without touching the file, treating the interaction as a guided tutorial. The instructional mode is the less-obvious use case, but it’s potentially more valuable for photographers who want to learn the underlying Photoshop workflow rather than delegate it entirely.
The web version adds an AI Markup tool that allows users to draw directly on the photo and attach a prompt to a specific region — a more spatially precise version of the same natural language interface.
The Model Layer
Adobe Firefly remains the default generation model, but the AI Assistant now integrates with 25+ external models, including Google’s Nano Banana 2, OpenAI’s image generation model, and Black Forest Labs’ Flux.2. Practically, this means the quality ceiling for specific tasks (photorealistic background replacement, complex object removal) is no longer limited to what Firefly alone can do. Adobe is positioning itself as the interface layer rather than the sole AI provider.
This is a meaningful architectural shift. Creative Cloud subscribers aren’t locked into Adobe’s AI quality — they’re getting access to a curated selection of the best available models for each specific task, mediated through Photoshop’s existing workflow.
What It Can’t Do Yet
The desktop exclusion is the biggest current limitation. The web and mobile apps cover casual and quick editing workflows, but professional photography editing happens predominantly in Photoshop desktop. Until the AI Assistant reaches the desktop version, its practical reach is constrained to lighter-use scenarios and mobile-first creators.
The beta status also means reliability and consistency are works in progress. For complex compositions, generated regions from multiple AI models will have different quality characteristics that a photographer may need to blend manually — the assistant reduces the labor but doesn’t eliminate craft judgment.
The So What
Adobe AI Assistant for Photoshop makes the natural language editing promise concrete for the first time on a professional-grade tool at scale. The 25+ model integration is more significant than the individual capabilities: Adobe is effectively becoming an AI editing orchestration layer, which means its competitive position isn’t dependent on any single AI model’s quality.
For working photographers already in Creative Cloud, the evaluation task is clear — test the public beta on representative edits from your actual workflow and measure time savings against the quality output. The desktop gap is real, but the web app is functional enough to assess whether the interface model changes how you think about editing iteration.
Recommended Reading:
Photo by Luca Sammarco on Pexels
Content created with AI assistance and reviewed for accuracy.
Join the conversation
Stack Insiders is our free community for readers who want to go deeper — share resources, ask questions, and connect with others across every vertical we cover.
Join Stack Insiders →