← Back to AI Tools | PCToolsOnline home

How to set up Stable Diffusion for the AI Tools page

Install a local Stable Diffusion WebUI on your own machine and connect it to PCToolsOnline to generate and edit images right in your browser. All processing happens on your GPU — nothing is uploaded.

1. Requirements

Component	Minimum	Recommended
GPU	4 GB VRAM (NVIDIA/AMD) or Apple M1	8+ GB VRAM (NVIDIA RTX 3060 or better) / M1 Pro
RAM	8 GB	16+ GB
Disk	10 GB free	30+ GB (each checkpoint is 2–7 GB)
OS	macOS 12+, Windows 10/11, Linux (Ubuntu 22+)
Python	Python 3.10. The installer will set it up if needed.

2. Choose your WebUI

PCToolsOnline talks to any server that speaks the A1111-compatible REST API. The two most popular choices:

WebUI	Repo	Notes
AUTOMATIC1111 WebUI	GitHub	Original, battle-tested, huge community. Good default choice.
Forge	GitHub	Fork of A1111 with speed optimizations. Lower VRAM usage. Same API.

The instructions below use A1111. For Forge, replace the repo URL and folder name — everything else is identical.

3. Install

macOS / Linux

# Clone the repo
git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git
cd stable-diffusion-webui

# First launch: it installs Python deps automatically (~5 min)
./webui.sh --api --cors-allow-origins="*" --listen

Windows

1. Install Python 3.10.6 (tick "Add to PATH").
2. Install Git.
3. Open a terminal and run:

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

4. Wait for the setup to finish (first launch downloads ~4 GB of dependencies).
5. Once the terminal shows Running on local URL: http://127.0.0.1:7860, close it and relaunch with the API flags (see next step).

4. Download a checkpoint

The WebUI needs at least one model file (a .safetensors checkpoint). Popular choices for beginners:

Model	Size	Style	Link
Stable Diffusion 1.5	~4.3 GB	General purpose, very compatible	Hugging Face
Stable Diffusion XL (SDXL)	~6.9 GB	Higher resolution (1024px), more detail	Hugging Face
Realistic Vision v6	~2 GB	Photorealistic portraits & scenes	CivitAI

Download the .safetensors file and place it in:

stable-diffusion-webui/models/Stable-diffusion/

5. Launch with the API and CORS enabled

macOS / Linux

cd stable-diffusion-webui
./webui.sh --api --cors-allow-origins="*" --listen

Windows

Edit webui-user.bat and change the COMMANDLINE_ARGS line:

set COMMANDLINE_ARGS=--api --cors-allow-origins=* --listen

Then double-click webui-user.bat to launch.

Production (restrict origin)

./webui.sh --api --cors-allow-origins="https://pctoolsonline.com" --listen

After startup you should see:

Running on local URL:  http://0.0.0.0:7860

Open http://localhost:7860 in a browser tab to verify the WebUI loads. You can use that UI independently, but our page only needs the API.

6. Connect from the AI Tools page

1. Open the AI Tools page and select the Generate (Stable Diffusion) tab.
2. Set the endpoint URL to http://localhost:7860.
3. Click Test connection. The model dropdown should list your checkpoints and the sampler dropdown should populate.
4. Switch to the Text → Image sub-tab.
5. Enter a prompt, e.g. a cat sitting on a desk, cinematic lighting, 4k.
6. Leave the default settings (512×512, 20 steps, CFG 7, Euler) and click Generate.
7. Wait 10–60 seconds. The generated image appears below. Click Download to save it as PNG.

7. Troubleshooting

"Cannot reach the endpoint" / network error

• Confirm the terminal still says Running on local URL.
• Verify you launched with --api and --cors-allow-origins. Without --api, the REST endpoints don't exist. Without CORS, the browser silently drops the response.
• Open http://localhost:7860/docs in a new tab. You should see the FastAPI docs page. If not, the WebUI isn't running or the port is wrong.

Model dropdown is empty after Test connection

• You haven't placed a checkpoint file in models/Stable-diffusion/. See step 4.
• If you just placed a file, click the refresh icon next to the checkpoint dropdown in the WebUI, or restart the WebUI.

Generation is very slow (CPU-only)

• If you see Using CPU in the terminal, the WebUI didn't detect a compatible GPU.
• For NVIDIA: install the latest CUDA toolkit and make sure nvidia-smi works.
• For AMD on Linux: install ROCm and launch with --use-rocm.
• For Mac M1/M2/M3: PyTorch MPS is used automatically. Make sure you use macOS 12.3+ and have Xcode CLT installed (xcode-select --install).

Out of memory (OOM)

• Add --medvram or --lowvram flags when launching the WebUI.
• Reduce the image size (e.g. 384×384 instead of 512×512).
• Use Forge instead of A1111 — it uses ~30% less VRAM for the same settings.

Output images look bad or random noise

• Increase steps to 30–50.
• Raise CFG scale to 9–12 for more prompt adherence.
• Add negative prompts: low quality, blurry, watermark, text.
• Try a different checkpoint — model choice matters more than any setting.

Quick reference: common launch commands

# Default (API + CORS + listen on all interfaces)
./webui.sh --api --cors-allow-origins="*" --listen

# Low VRAM (4 GB GPU)
./webui.sh --api --cors-allow-origins="*" --listen --medvram

# Very low VRAM (3 GB GPU)
./webui.sh --api --cors-allow-origins="*" --listen --lowvram

# SDXL with xformers (faster, needs xformers installed)
./webui.sh --api --cors-allow-origins="*" --listen --xformers

What's next?

• Browse CivitAI for more checkpoints and LoRA add-ons.
• Try the Image → Image mode to restyle existing photos.
• Interested in image understanding (alt text, OCR, tags) instead of generation? See the Ollama local setup tutorial.