ABrush

The friction of switching between Photoshop and web AI tools

The context-switch tax of toggling between Photoshop and a browser-based AI generator is not just friction — it’s where layer state gets lost, masks get misaligned, and iteration grinds to a halt. ABrush plugs generation directly into Photoshop so the canvas you’re editing and the model you’re prompting are the same window.

How it works

Inside the plugin, you run Stable Diffusion, Flux, and other supported models without leaving Adobe’s workspace — generating variations, inpainting masked regions, and refining outputs by hand on the same layer stack. Generation history and shareable presets let you return to a working prompt-and-settings combination without reconstructing it from memory. The vendor states no image data is used for model training, which matters for commercial studios with IP concerns.

Limitations

The ceiling appears when you need API access for external pipelines or self-hosted deployment — neither is available.

Pros

Generation runs inside Photoshop on the active document, so masks, layer order, and adjustment layers stay intact across the AI step — no export-reimport cycle that breaks composite alignment. ControlNet and IP-Adapter support inside the plugin, which means reference-image constraints work without switching to a separate ComfyUI or Automatic1111 instance and re-importing the result. Generation history and preset sharing across a team, so a prompt-and-settings combination that worked on Monday is retrievable and shareable rather than reconstructed from memory each session. The vendor states image data is not used for training.

Cons

No API is available, so any pipeline that needs to trigger generation programmatically — batch asset production, downstream CMS publishing, or CI-integrated design systems — has no integration path. Generation requires an active internet connection with no local inference option, which means studios in air-gapped environments or with strict data-residency rules cannot use ABrush regardless of the no-training-data policy.

Who it is for / who should skip it

Best for digital artists using Photoshop, controlled AI-assisted image editing, iterative creative workflows, and commercial projects requiring data privacy. Teams that need API access or local inference should skip it.