● LIVEReading: Run Stable Diffusion Locally with a Pro UI (InvokeAI)Total time: 12 minSteps: 5Worked first time: 72%● LIVEReading: Run Stable Diffusion Locally with a Pro UI (InvokeAI)Total time: 12 minSteps: 5Worked first time: 72%
CBW
Mediumgithub.com/invoke-ai/InvokeAI2026-05-19

Run Stable Diffusion Locally with a Pro UI (InvokeAI)

Install InvokeAI on your own computer and generate images with Stable Diffusion, FLUX, and more — no cloud subscription, no coding required. Everything runs through a polished browser-based interface.

// Build stats

  • Total time12 min
  • Number of steps5
  • DifficultyMedium
  • Worked first time72%
// Before you start

What you need

  • Windows 10/11, macOS 12+, or Linux
  • NVIDIA GPU with 6 GB+ VRAM (AMD works on Linux; Apple Silicon works via MPS)
  • At least 20 GB free disk space for models
  • Stable internet connection for the initial download
  • No coding experience needed
01
Step 1 of 5

Download and open the InvokeAI Launcher

3 min

InvokeAI installs through a small desktop app called the Launcher — it handles Python, dependencies, and updates for you, so you never touch a terminal or install Python yourself. The Launcher lives in its own GitHub repository, separate from the main InvokeAI code. Download it, then get the app open.

Terminal · mac
$ # 1. Open this URL and download the file for your OS:
$ https://github.com/invoke-ai/launcher/releases/latest
$ # Windows -> Invoke.Community.Edition.Setup.latest.exe
$ # macOS -> Invoke.Community.Edition-latest-arm64.dmg
$ # Linux -> Invoke.Community.Edition-latest.AppImage
$
$ # 2. Open it:
$ # Windows: run the .exe — it installs the Launcher and makes a desktop shortcut
$ # macOS: open the .dmg, drag the Invoke app to Applications, then launch it
$ # Linux: make the .AppImage executable (right-click > Properties), then run it
What you should see
The InvokeAI Launcher window opens, showing an Install button.
This might happen

Windows SmartScreen or macOS 'unidentified developer' blocks the file.

Windows: click 'More info' then 'Run anyway'. macOS: System Settings > Privacy & Security > 'Open Anyway'. This is the official Launcher from the invoke-ai/launcher GitHub releases page.

02
Step 2 of 5

Install InvokeAI through the Launcher

10-20 min

Inside the Launcher, click the Install button and follow the prompts. The Launcher downloads everything InvokeAI needs — Python, PyTorch, and several gigabytes of dependencies. This is the longest step because of the download size; you can leave it running and come back. (If you ever reinstall, the Launcher can also detect and update an existing installation.)

Terminal · mac
$ # No terminal command — do this inside the Launcher app:
$ # 1. Click the 'Install' button
$ # 2. Follow the on-screen prompts (choose an install location if asked)
$ # 3. Wait for the download + install to finish
What you should see
The Launcher shows installation progress, then InvokeAI appears as ready to launch.
This might happen

Installation stalls or errors out mid-way.

Check your internet connection. On Windows, temporarily disable antivirus real-time scanning — it sometimes blocks file writes — then click Install again; it resumes where it left off.

03
Step 3 of 5

Launch InvokeAI and open the web UI

2 min

From the Launcher, start InvokeAI. It spins up a local web server on your machine and opens the full InvokeAI interface in your browser. From this point on, no internet connection is required.

Terminal · mac
$ # In the Launcher app, click the 'Launch' button.
$ # When it is ready, your browser opens InvokeAI automatically.
$ # If it does not, open this address manually:
$ http://127.0.0.1:9090
What you should see
The InvokeAI interface loads in your browser at http://127.0.0.1:9090.
This might happen

The browser opens but shows a blank page or 'connection refused'.

Wait 30 more seconds — the server needs time to finish loading — then refresh. If it is still blank, check the Launcher's log output for error messages.

04
Step 4 of 5

Download your first image-generation model

5-30 min depending on model size

InvokeAI needs at least one AI model to generate images. Inside the web UI, click the 'Model Manager' icon (looks like a stack of layers, usually in the left sidebar). From there you can browse and install starter models with one click. SDXL is a good all-around starting point. FLUX models produce higher quality but are larger (12+ GB). Pick one that fits your disk space and VRAM.

Terminal · mac
$ # No command needed — do this entirely inside the browser UI:
$ # 1. Open http://127.0.0.1:9090
$ # 2. Click the Model Manager icon in the left sidebar
$ # 3. Click 'Add Model' or browse the Starter Models tab
$ # 4. Select a model (e.g. SDXL 1.0) and click Install
$ # 5. Wait for the download progress bar to complete
What you should see
The model appears in your model list with a green checkmark or 'Ready' status.
This might happen

Model download fails or shows an error about Hugging Face access.

Some models (especially FLUX) require a free Hugging Face account and accepting a license. Create an account at huggingface.co, accept the model's license on its page, generate an access token, and paste it into InvokeAI's settings under 'HuggingFace Token'.

05
Step 5 of 5

Generate your first image

2-5 min

Go to the main Generate tab. Type a description of what you want to see in the Prompt box — be specific and descriptive. Select your installed model from the model dropdown at the top. Leave all other settings at their defaults for now. Click the Generate button and watch the image appear. Generation time depends on your GPU — expect 5 to 60 seconds per image.

Terminal · mac
$ # No command needed — do this in the browser UI:
$ # 1. Click the 'Generate' tab (canvas icon) in the left sidebar
$ # 2. Type a prompt, e.g.:
$ # a photorealistic red fox sitting in a snowy forest, golden hour lighting
$ # 3. Confirm your model is selected in the top dropdown
$ # 4. Click the purple 'Generate' button
$ # 5. Your image appears in the gallery on the right when done
What you should see
A generated image matching your prompt appears in the gallery panel. Metadata (prompt, model, seed) is saved automatically.
This might happen

Generation crashes or you see 'CUDA out of memory'.

Your GPU does not have enough VRAM for the current settings. In the settings panel, reduce image size to 512x512 (for SD 1.5) or 1024x1024 (for SDXL), and enable 'xFormers' or 'Attention Slicing' in the settings if available.

// Status

cooked. baked. worked.

A fully local, browser-based AI image generation studio running on your own hardware. You can generate, edit, inpaint, and upscale images without any cloud service or ongoing subscription cost.

// the honest bit

The honest part

InvokeAI is powerful but hardware-dependent. Without a dedicated GPU you will get very slow generation times (minutes per image on CPU). FLUX models require 12+ GB VRAM and 20+ GB disk space — they will not run on budget GPUs. Some premium models (FLUX Dev, SD 3.5) require accepting Hugging Face license agreements. The Launcher simplifies setup greatly, but occasional dependency issues do occur — the InvokeAI Discord is active and helpful when things break.