The Complete Guide to Stable Diffusion and Forge: Unrestricted AI Image Generation on Your Own Computer

What Is Stable Diffusion and Why Run It Locally?

Stable Diffusion is a free, open-source AI image generator that runs entirely on your own computer. Unlike cloud-based tools like Midjourney, DALL-E, or Adobe Firefly, running Stable Diffusion locally means:

No content filtering — full creative freedom with zero prompt restrictions
No subscriptions — completely free after initial setup
No usage limits — generate as many images as you want, 24/7
Total privacy — your prompts and images never leave your machine
Offline capable — works without an internet connection once installed

This guide covers everything you need to know: the two best interfaces (Automatic1111 and Forge), how to install them on Windows and Linux, how to enter prompts, which models work with your GPU, and how to avoid the content restrictions that cloud tools enforce.

Automatic1111 vs Forge: Which Interface Should You Use?

Stable Diffusion is the AI engine that generates images. To actually use it, you need a web interface (UI) that lets you type prompts and see results. The two most popular options are:

Automatic1111 (A1111)

The original and most widely used Stable Diffusion interface. It has the largest community, the most extensions, and years of documentation and tutorials.

Forge (SD WebUI Forge)

A newer fork of Automatic1111 created by lllyasviel (the developer behind ControlNet). It looks and works nearly identically to A1111 but is significantly faster and uses less GPU memory.

Feature	Automatic1111	Forge
Content Filtering	None	None
Prompt Restrictions	None	None
Generation Speed	Baseline	30-75% faster
VRAM Usage	Higher	Significantly lower
UI/Interface	Standard tabbed layout	Nearly identical to A1111
Extension Support	Huge library (thousands)	Compatible with most A1111 extensions
Community Size	Largest	Growing fast
Best For	Maximum compatibility, most tutorials	Speed, low VRAM GPUs (6-8GB)

Our recommendation: If you’re starting fresh, go with Forge. You get the same experience but faster performance and lower memory requirements. If you already have A1111 set up with extensions you rely on, there’s no rush to switch.

Content Filtering: Local vs Cloud Tools

One of the biggest reasons people run Stable Diffusion locally is creative freedom. Here’s how content filtering compares across platforms:

Cloud Tools (Filtered)

Midjourney — Heavy content filtering, rejects many prompts, bans accounts for violations
DALL-E 3 / ChatGPT — Strict filters, blocks a wide range of prompt keywords
Adobe Firefly — Strict commercial-safe filtering
Bing Image Creator — Same DALL-E filters as ChatGPT
Leonardo AI — Has a content filter toggle but still restricts certain content

Local Tools (Unfiltered)

Automatic1111 — No content filtering whatsoever
Forge — No content filtering whatsoever
ComfyUI — No content filtering whatsoever

Why is there no filter? These are open-source tools that run 100% on your hardware. There is no company server in the middle to enforce rules. You download the model, you run it locally, and nothing is censored, blocked, or reported.

Note: Some models shipped by Stability AI included a basic “safety checker” module, but it can be easily disabled in settings. Most community models on platforms like CivitAI do not include any safety checker at all.

How to Disable the Safety Checker (If Present)

In Automatic1111 or Forge, if you encounter a black image with a safety warning:

Go to Settings (tab at the top)
Search for “NSFW” or “safety checker”
Uncheck or disable the safety checker option
Click Apply Settings and Reload UI

Alternatively, launch with the command line flag: --disable-safe-unpickle

Most users never encounter this since community models don’t include a safety checker by default.

How to Check If You Already Have Stable Diffusion Installed

On Windows

Open PowerShell or Command Prompt and run:

dir C:\stable-diffusion-webui
dir %USERPROFILE%\stable-diffusion-webui

For Forge:

dir C:\stable-diffusion-webui-forge
dir %USERPROFILE%\stable-diffusion-webui-forge

If the folder exists, you have it installed.

On Linux

Open a terminal and run:

ls ~/stable-diffusion-webui
ls ~/stable-diffusion-webui-forge

Or search for it:

find ~ -maxdepth 3 -name "webui.sh" 2>/dev/null

Check Your Version

Once you find the installation folder:

cd ~/stable-diffusion-webui
git log --oneline -1
git describe --tags

This shows your exact version and commit. You can also see the version in the bottom left corner of the WebUI when it’s running in your browser.

Installation Guide: Windows

Installing Automatic1111 on Windows

Install Python 3.10.x from python.org (check “Add to PATH” during install)
Install Git from git-scm.com

Clone the repository:

git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git
cd stable-diffusion-webui

Launch:
```
webui-user.bat
```
Wait for the first-time setup (downloads dependencies and a base model automatically)
Open your browser to http://localhost:7860

Installing Forge on Windows

Download the one-click installer from the Forge GitHub releases page (search “stable-diffusion-webui-forge” on GitHub)
Extract the zip file to a folder
Run: run.bat
Open your browser to http://localhost:7860

Installation Guide: Linux

Installing Automatic1111 on Linux

sudo apt install python3 python3-venv python3-pip git -y
git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git
cd stable-diffusion-webui
bash webui.sh

The script handles everything — creates a virtual environment, installs PyTorch, and downloads a base model. First launch takes 10-20 minutes.

Installing Forge on Linux

git clone https://github.com/lllyasviel/stable-diffusion-webui-forge.git
cd stable-diffusion-webui-forge
bash webui.sh

Both interfaces open at http://localhost:7860 by default. To access from another device on your network, launch with: --listen --share

How to Enter Prompts

Once the WebUI is running in your browser, generating images is simple:

Step 1: Write Your Positive Prompt

The top text box is your positive prompt — describe what you want to see:

a cyberpunk city at night, neon lights reflecting on wet streets, rain, photorealistic, highly detailed, 8k resolution, cinematic lighting

Step 2: Write Your Negative Prompt

The bottom text box is your negative prompt — describe what you do NOT want:

blurry, low quality, deformed, extra fingers, mutated hands, watermark, text, logo, bad anatomy, ugly, distorted

Step 3: Adjust Settings

Sampling Steps: 20-30 (more = more detail, but slower)
CFG Scale: 7-12 (higher = follows prompt more strictly)
Sampler: DPM++ 2M Karras (great default for most models)
Resolution: 512×512 for SD 1.5, 1024×1024 for SDXL models

Step 4: Click Generate

Hit the Generate button and wait a few seconds to a minute depending on your GPU. Your image appears in the output panel.

Prompt Tips for Better Results

Be specific: “A golden retriever playing in autumn leaves, soft afternoon sunlight, Canon EOS R5” beats “dog in park”
Add quality boosters: Include terms like “masterpiece, best quality, highly detailed, professional” in your positive prompt
Use emphasis: Put parentheses around important words to increase their weight: (neon lights:1.3) makes neon lights 30% more prominent
Specify art styles: “oil painting,” “watercolor,” “anime,” “photorealistic,” “3D render,” “pencil sketch”
Describe lighting: “golden hour,” “dramatic shadows,” “studio lighting,” “backlit,” “volumetric fog”
Set the camera: “close-up portrait,” “wide angle landscape,” “bird’s eye view,” “macro photography”

Models by GPU VRAM: What Can Your Hardware Run?

Different Stable Diffusion models require different amounts of GPU memory (VRAM). Here’s a practical guide to matching models with your hardware:

4GB VRAM (GTX 1650, GTX 1050 Ti)

Recommended UI: Forge (critical for low VRAM — A1111 may struggle)
Best models: SD 1.5 based models at 512×512
Popular models: Realistic Vision 1.5, DreamShaper 1.5, Anything V5 (anime)
Launch flag: --medvram-sdxl --xformers
What to expect: Functional but slow. Stick to 512×512 resolution. Forge makes this tier much more usable.

6GB VRAM (RTX 2060, RTX 3060 Mobile, GTX 1660)

Recommended UI: Forge or A1111
Best models: SD 1.5 (comfortable), SDXL (possible with Forge)
Popular models: Realistic Vision, DreamShaper XL, Juggernaut XL
Launch flag: --medvram --xformers
What to expect: SD 1.5 runs great. SDXL works in Forge with optimizations. Can do 768×768 comfortably.

8GB VRAM (RTX 3060 Ti, RTX 3070, RTX 4060)

Recommended UI: Either A1111 or Forge
Best models: SD 1.5 (fast), SDXL (comfortable), Flux Schnell (with Forge)
Popular models: Juggernaut XL, RealVisXL, SDXL base, Flux Schnell
What to expect: The sweet spot for most users. SDXL at 1024×1024 runs well. Can use ControlNet and other advanced features.

12GB VRAM (RTX 3060 12GB, RTX 4070, RTX 4080)

Recommended UI: Either — both run smoothly
Best models: SDXL (fast), Flux Dev, Flux Schnell (fast), large LoRA stacks
Popular models: Flux Dev, Juggernaut XL, RealVisXL, Pony Diffusion XL
What to expect: Runs everything comfortably. Can stack multiple LoRAs, use ControlNet, and generate at high resolutions. Flux Dev works great here.

16-24GB VRAM (RTX 4080, RTX 4090, RTX 3090)

Recommended UI: Either — go wild
Best models: Everything. Flux Pro, Flux Dev (fast), SDXL (instant), large batches
What to expect: No limitations. Generate in seconds. Run multiple models. Use every feature at maximum quality.

AMD GPU Users

Linux: AMD GPUs work via ROCm. Install ROCm drivers, then launch with --use-rocm
Windows: Limited support via DirectML. Launch with --use-directml
Performance: Expect roughly 50-70% of equivalent NVIDIA performance

Mac (Apple Silicon) Users

M1/M2/M3/M4 chips work with Stable Diffusion via the MPS backend
Launch with: --use-mps (usually auto-detected)
8GB unified memory handles SD 1.5 well. 16GB+ handles SDXL.
Slower than equivalent NVIDIA GPUs but fully functional

Where to Download Models

Models are the “brains” of Stable Diffusion. Different models produce different styles:

CivitAI.com — The largest model library. Browse by category, style, and popularity. Community reviews and example images for every model.
Hugging Face — The official home for many models. More technical but reliable.

How to Install a Model

Download the model file (.safetensors format — always prefer this over .ckpt for security)
Place it in your stable-diffusion-webui/models/Stable-diffusion/ folder
Click the refresh button next to the model dropdown in the WebUI
Select your new model from the dropdown

Recommended Starter Models

Model	Style	Base	Min VRAM
Realistic Vision	Photorealistic	SD 1.5	4GB
DreamShaper	Artistic/Fantasy	SD 1.5	4GB
Anything V5	Anime/Illustration	SD 1.5	4GB
Juggernaut XL	Photorealistic	SDXL	6GB
RealVisXL	Photorealistic	SDXL	6GB
DreamShaper XL	Artistic	SDXL	6GB
Pony Diffusion XL	Versatile/Stylized	SDXL	8GB
Flux Schnell	High quality general	Flux	8GB
Flux Dev	Highest quality	Flux	12GB

Troubleshooting Common Issues

“CUDA out of memory” Error

Switch to Forge (uses less VRAM)
Lower your resolution (try 512×512)
Add --medvram or --lowvram to your launch flags
Close other GPU-intensive apps (games, video editors, other AI tools)

Black Images

The safety checker may be enabled — disable it in Settings
Try adding --no-half to your launch flags (fixes some GPU issues)
Switch to a different sampler (try “Euler a” or “DPM++ 2M Karras”)

Slow Generation

Enable xformers: add --xformers to launch flags
Switch to Forge for 30-75% speed improvement
Reduce steps to 20 (diminishing returns after 30)
Generate at 512×512 and upscale after

WebUI Won’t Start

Make sure Python 3.10.x is installed (not 3.12+, which can cause issues)
On Linux, ensure NVIDIA drivers and CUDA are installed: nvidia-smi
Delete the venv folder and relaunch to force a fresh install of dependencies

Bottom Line

Running Stable Diffusion locally with Automatic1111 or Forge gives you the most powerful, unrestricted AI image generation available. No content filters, no subscriptions, no limits. Forge is the better choice for most people in 2026 — it’s faster, uses less memory, and works identically to A1111. Match your model choice to your GPU’s VRAM, start with the recommended models above, and you’ll be generating professional-quality AI images in minutes.