The XDA piece on Krita’s AI plugins matching Photoshop’s $120-a-year tools caught a wave that has been building all year. Local image generation on consumer GPUs got faster, the model ecosystem opened up, and free apps started doing what subscription services charged for. We’ve spent the past few months bouncing between every serious tool for desktop AI image generation. These are the seven that earned their slot on a Windows, macOS, or Linux machine.
What to look for in a desktop AI image generator
Most of the noise in this space is product positioning. The choices that matter are simpler:
- Local vs cloud. Local apps run on your GPU, no API costs, no upload of source images. Cloud apps shift the compute and the privacy question.
- Model freedom. Some apps lock you to a single model family. Others run Stable Diffusion XL, Flux, SD3, and community LoRAs side by side.
- Inpainting and outpainting depth. Generating an image is the easy part. The good apps let you mask, edit, and extend an image without going round-trip through a separate editor.
- Node graphs vs forms. Beginners want forms; advanced users want node graphs (ComfyUI) that let them build pipelines.
- Hardware floor. A 6GB VRAM card can run quantized SDXL and Flux. Below that, prefer macOS Apple Silicon paths or cloud.
- Commercial license. Some models forbid commercial output. Adobe and Stability cleared this; community LoRAs are mixed.
Quick comparison
| App | Best for | Platforms | Free | Starting price | Hardware floor |
|---|---|---|---|---|---|
| Stable Diffusion WebUI (Forge / A1111) | Power users on Nvidia | Windows, Linux, macOS | Yes | Free | 6GB VRAM |
| ComfyUI | Node-graph workflows | Windows, Linux, macOS | Yes | Free | 6GB VRAM |
| Fooocus | Beginner-friendly local | Windows, Linux | Yes | Free | 4GB VRAM |
| InvokeAI | Studio workflows | Windows, Linux, macOS | Yes | Enterprise plans on request | 8GB VRAM |
| Krita with AI Diffusion plugin | Painters who need generative fill | Windows, Linux, macOS | Yes | Free | 6GB VRAM |
| Adobe Photoshop (Firefly) | Pros who need commercial licensing | Windows, macOS | Trial | $22.99/mo | Modest CPU/GPU |
| DiffusionBee | macOS one-click installs | macOS | Yes | Free | M1 or newer |
| Draw Things | Apple Silicon and iPad parity | macOS, iOS | Yes | Optional Pro tier | M1 or newer |
The apps
1. Stable Diffusion WebUI (Forge / A1111) — Best for Nvidia power users
Stable Diffusion WebUI in its Forge and Automatic1111 forks is still the default for local image generation on Nvidia GPUs. It runs Stable Diffusion XL, SD3, Flux, and most community models out of the box, with ControlNet, LoRAs, and inpainting baked in. The 2026 Forge releases brought meaningful speedups on Ada and Blackwell cards.
Where it falls short: Setup is Python-flavored even with one-click installers. UI feels dated next to InvokeAI or ComfyUI. AMD support works through DirectML but isn’t first-class.
Pricing:
- Free: open source
- Paid: none
- Platforms: Windows, Linux, macOS (slower)
Download: Stable Diffusion WebUI Forge on GitHub
Bottom line: Pick this when you have an Nvidia card and you want maximum flexibility.
2. ComfyUI — Best node-graph workflow
ComfyUI wraps generative pipelines as a node graph. Every step (text encoding, sampling, latent upscale, ControlNet, post-processing) is a node you wire together. The 2026 Manager addon hides the messy install bits and bundles the popular custom node packs.
Where it falls short: Visual graph approach overwhelms beginners. Sharing workflows means sharing JSON, which sometimes breaks across versions. Less hand-holding than Fooocus.
Pricing:
- Free: open source
- Paid: ComfyUI Cloud for hosted runs, billed per minute
- Platforms: Windows, Linux, macOS
Download: ComfyUI
Bottom line: Pick this when you want repeatable pipelines, batch jobs, or to ship a workflow to a team.
3. Fooocus — Best beginner-friendly local
Fooocus strips Stable Diffusion XL down to a clean form: type a prompt, pick a style, press Generate. The defaults are tuned, the install is a single download, and the 4GB VRAM floor opens the door to entry-level GPUs.
Where it falls short: Less flexibility than WebUI. Custom models and LoRAs are supported but not surfaced cleanly. macOS support trails Windows.
Pricing:
- Free: open source
- Paid: none
- Platforms: Windows, Linux
Download: Fooocus on GitHub
Bottom line: Pick this when you’re new to local generation and you want quick, attractive results.
4. InvokeAI — Best studio workflow
InvokeAI is the closest thing to a Photoshop-grade local generative studio. Canvas-based inpainting, layer-aware masks, and a Unified Canvas that handles outpainting feel like real image editing rather than prompt-and-pray.
Where it falls short: Higher VRAM floor than Fooocus. Pro features push toward a subscription. Smaller LoRA ecosystem than WebUI.
Pricing:
- Free: open-source Community edition
- Paid: Invoke Pro and Enterprise quotes on request
- Platforms: Windows, Linux, macOS
Download: InvokeAI
Bottom line: Pick this when canvas editing matters and you want the closest thing to a Photoshop-like generative workflow.
5. Krita with AI Diffusion plugin — Best for painters
Krita plus the AI Diffusion plugin is the combination XDA’s piece highlighted: the brush-first painting app with a generative fill that runs on your own GPU. The plugin handles inpainting, control modes, and live painting modes. It’s free, local, and inside the canvas you already paint in.
Where it falls short: Plugin install still needs Python wrangling on Windows. Performance trails dedicated generators on the same card. Some models require manual download.
Pricing:
- Free: Krita and the plugin
- Paid: none
- Platforms: Windows, Linux, macOS
Download: Krita on Steam
Bottom line: Pick this when generative fill needs to sit inside a real painting app.
6. Adobe Photoshop (Firefly) — Best for commercial-safe work
Adobe Photoshop with Firefly is the choice when commercial licensing matters. Adobe trained Firefly on licensed and public-domain content and indemnifies enterprise customers. Generative Fill, Generative Expand, and the new Image 3 model handle the same brief as Krita’s plugin with corporate guardrails.
Where it falls short: Subscription only. Generation runs through Adobe’s cloud, so source images leave your machine. Style range narrower than open models for non-Western art.
Pricing:
- Free: 7-day trial
- Paid: $22.99/month standalone, $59.99/month Creative Cloud All Apps
- Platforms: Windows, macOS
Download: Adobe Photoshop
Bottom line: Pick this when client legal teams insist on commercial-safe AI output.
7. DiffusionBee — Best for Mac one-click
DiffusionBee is the Apple Silicon native that wraps Stable Diffusion behind a one-click install for macOS. It runs Stable Diffusion XL on M1 through M4 cleanly and adds a focused inpainting view. No Terminal, no Python virtualenv.
Where it falls short: Limited custom node support. Generation speed trails Nvidia desktops. Smaller model selection than WebUI.
Pricing:
- Free: open source
- Paid: none
- Platforms: macOS
Download: DiffusionBee
Bottom line: Pick this when you’re on a Mac and you don’t want to fight Python.
8. Draw Things — Best for Apple Silicon and iPad parity
Draw Things is the model-flexible Apple Silicon app that mirrors the same workflow across macOS and iPadOS. The 2026 build added Flux support, LoRA chaining, and a clean control net panel. It’s the closest thing to ComfyUI’s flexibility on Apple hardware.
Where it falls short: Mac-and-iPad only. Sharing pipelines across devices is fine, sharing across platforms is not. Power features sit behind menu dives.
Pricing:
- Free: full feature set
- Paid: optional Pro support tier
- Platforms: macOS, iOS
Download: Draw Things
Bottom line: Pick this when you split work between a Mac and an iPad and you want the same models on both.
How to pick the right one
If you want the simplest local install on Windows, pick Fooocus. Four-gigabyte VRAM floor and sane defaults.
If you want maximum flexibility on Nvidia, pick Stable Diffusion WebUI Forge. Every model, every LoRA, every ControlNet, all in one place.
If you want repeatable pipelines, pick ComfyUI. The node graph pays off as soon as you build the second version of a workflow.
If you want canvas-based editing inside the generator, pick InvokeAI. The Unified Canvas is the right answer for retouching workflows.
If you already paint, pick Krita with the AI Diffusion plugin. Generative tools inside the brush app you already learned.
If commercial licensing matters, pick Photoshop. Firefly’s indemnification matters for client work.
If you live on Apple Silicon, pick DiffusionBee for one-click installs or Draw Things if you want power features without giving up Mac-native UX.
FAQ
What is the best free AI image generator for desktop? Stable Diffusion WebUI Forge for Nvidia users who want maximum control. Fooocus for beginners. DiffusionBee or Draw Things for Apple Silicon.
Can I run AI image generators on a laptop? Yes, with caveats. Fooocus runs on 4GB VRAM. Apple Silicon Macs run DiffusionBee and Draw Things efficiently. Anything below 6GB VRAM on Nvidia means longer generations and limited model choice.
Do I need an Nvidia GPU for local AI image generation? No, but Nvidia is the fastest path on Windows and Linux. AMD users can run WebUI via DirectML or ROCm with worse performance. Apple Silicon runs MPS-accelerated apps natively.
Is using AI image generators legal for commercial work? Generally yes, but it depends on the model. Adobe Firefly is licensed for commercial output and indemnifies enterprise users. Open Stable Diffusion outputs are fine for most uses; LoRAs and fine-tunes vary in their licenses.
What is the best AI image generator for inpainting? InvokeAI’s Unified Canvas, Photoshop’s Generative Fill, and Krita with AI Diffusion all do this well. Choose based on whether you want local, commercial-safe, or paint-first workflow.
Which app supports Flux and SD3? ComfyUI, WebUI Forge, InvokeAI, and Draw Things all run Flux and SD3. Fooocus focuses on SDXL. DiffusionBee added SDXL Turbo support in 2025 and is catching up on newer models.