ABrush Review — AI Image Generation Inside Adobe Photoshop

What Is ABrush?

ABrush is a Photoshop-native AI studio that runs inside Adobe Photoshop, so artists generate, edit, and iterate on images without leaving the open document. It grew out of three years of internal tooling at AB Games, the Ukrainian-rooted studio behind the long-running hidden-object game Hidden City, before spinning out as a standalone product. The commercial operator is Cyprus-registered PERLTEC Ltd, and the intended audience is individual artists plus small to mid-sized studios who already live inside Photoshop.

The product is in active development. Access is staged — the team runs time-limited, free testing rounds for selected candidates and is preparing an Early Access program, with a global launch planned for autumn 2026. Because all complex computations happen in the cloud, the plugin needs no local graphics card and keeps results flowing between the active document and the model provider.

Key Features

Instead of a single engine, the plugin bundles more than 20 top image models — the site lists 23+ — spanning families such as Stable Diffusion, Flux, and the team's own Gwen Image. Recent test builds added engines like Flux.2 Pro, Nano Banana 2, and Seedream, and the available model list adapts to whatever work mode is active.

Five documented modes cover the pipeline: Txt2Img for from-scratch generation, Edit for text-driven changes to an existing image, Upscale to raise resolution, Img2Img to redraw a visual reference, and Action modes for batch or scripted operations. A prompt card adds translation, prompt improvement, and image description, while ControlNet and IP-Adapter references give structural control over composition and style.

Window layouts stay modular: a main window holds mode, model, and prompt controls, a workflow window groups generation tools and presets, and a gallery window keeps history for reuse. The interface lets you switch models without losing the prompt, which keeps style tests cheap during a production day.

• 23+ bundled models including Stable Diffusion, Flux, and Gwen Image
• Five modes: Txt2Img, Edit, Upscale, Img2Img, and Action
• ControlNet and IP-Adapter references for guided output
• Prompt translation, improvement, and image description
• Background removal plus mask-based local edits

How to Use It

Install the plugin into Photoshop, open a document, and select the area you want to populate. Type a prompt or load a saved preset, choose a model and mode, and the server returns one to four variations in the canvas area while the job runs in the cloud.

Keep what works and refine the rest with Photoshop's native layers, masks, and brushes, because the artist stays the final decision-maker at every step. Strong results are reusable through generation history and shareable presets, and base resolution sits at 1024px or 2048px depending on the model, with the Upscale mode extending it further.

If the standard Creative Cloud install fails, the plugin can be installed manually from the .ccx package — a documented fallback for managed machines. The settings window also controls login, favorites lists, prompt save behavior, and multi-prompting with a separator character for batch generation.

• Install via Creative Cloud, select an area, and prompt
• Generate one to four variations directly in the document
• Refine with layers, masks, and Photoshop brushes
• Reuse strong results via history and shareable presets

Why Game-Art Teams Care

The tool was built inside a game studio, which shows in the workflow. AB Games has tested the plugin with 52 artists generating more than 50,000 images a month, and the product promise is to automate repetitive production work so artists can spend time on the parts only they can do.

Style consistency is the other production win. Shared presets and generation history let an art team lock a visual direction across dozens of assets, while ControlNet conditions keep structure, poses, and layout under control. Concept art, character design, environment work, and background generation are the natural starting points.

The founding argument, per the team's own posts, is that artists live inside Photoshop while AI tools live elsewhere — so the product removes the export-import-browser loop between them.

• Born from three years of internal use at AB Games
• Tested with 52 artists generating 50,000+ images monthly
• Presets keep style consistent across an art team
• Tuned for concept art, characters, environments, and backgrounds

How It Compares to Adobe Firefly

ABrush and Adobe Firefly both live in Adobe's ecosystem, but they are different products. Firefly is Adobe's curated generative model family, integrated into Creative Cloud apps and governed by Adobe's own terms; ABrush is a third-party studio that aggregates many models inside Photoshop and exposes controls such as ControlNet and IP-Adapter that Firefly does not.

The trade-off is distribution of control. Because generation runs through third-party cloud model providers, commercial rights and data handling follow the terms of this plugin rather than Adobe's, so studios should read those terms before piping confidential assets through the pipeline.

Standalone Alternatives

Midjourney is the aesthetic leader for standalone prompting, while Adobe Firefly stays inside the Creative Cloud family that Photoshop users already pay for. Leonardo AI offers a polished browser canvas with sharp controls and an API, and Krea AI covers real-time generation with a large model hub in one subscription.

For artists who prefer open tooling, Stable Diffusion and its many interfaces run locally with full model control. Each of these is a web or desktop app rather than an in-document plugin, so for Photoshop-first artists the deciding factor is usually whether generation should live inside the canvas or in a separate suite.

• Adobe Firefly — integrated with Creative Cloud apps
• Midjourney — leading aesthetic quality and community
• Leonardo AI — polished web canvas with controls and API
• Krea AI — real-time generation and a large model hub
• Stable Diffusion — local and open-source generation

Data, Ownership, and Requirements

The official FAQ is explicit on the basics: all generated images belong to you, you keep full rights to use them commercially without restriction, and the team does not train neural networks on your input images or results without your direct consent. The terms of use repeat that your local or cloud content will not be used to train generative AI models, and local content is never reviewed.

Because generation runs on the platform's cloud and its model providers, prompts and source material are transmitted off-device, and the terms note data is processed in the United States. Since computation is server-side, the only machine requirement is Adobe Photoshop 26.1 or newer plus an always-on internet connection — no powerful GPU is needed.

• You own generated images with full commercial rights
• No training on your images or results without consent
• Cloud compute in the US; prompts travel to model providers
• Requires Photoshop 26.1+ and an active internet connection
• No graphics card required

ABrush Pricing

ABrush is in staged testing — free, time-limited test rounds plus an upcoming Early Access program. The team has confirmed a permanent free tier, with a subscription system (Lite, Pro, Premium, Teams) planned for the autumn 2026 launch. Public dollar pricing was not verified.

Best For

Recommended use cases and scenarios where ABrush shines.

Frequently Asked Questions

Common questions about ABrush, answered.