The single step that trips up almost everyone generating their first Stable Diffusion image locally is the model file. You can install the interface perfectly and still stare at an empty dropdown, because Stable Diffusion needs a checkpoint dropped into a specific folder before a single prompt will produce anything. Get the checkpoint in the right place and the rest is genuinely a few clicks: type a prompt, press generate, watch your gaming PC turn out a 512x512 image in seconds.

Quick Answer

To generate your first local Stable Diffusion image, install a WebUI such as AUTOMATIC1111, download a checkpoint file, and place it in the models/Stable-diffusion folder. Restart the WebUI, pick the model from the dropdown, type a prompt and click Generate. On a 12GB RTX GPU a first 512x512 image takes only a few seconds once the model is loaded.

What you need before you start

  1. A gaming PC with a capable NVIDIA GPU. Around 12GB of VRAM is a comfortable starting point for 512x512 generation with room to grow.
  2. Python 3.10.6 specifically, not the newest release, since the WebUI is fussy about versions. Tick "Add Python to PATH" during install.
  3. Git for Windows, used to download and update the WebUI.
  4. A safetensors checkpoint file, which is the actual Stable Diffusion model. Beginner-friendly models like DreamShaper or Realistic Vision are good first choices. Use safetensors rather than older ckpt files from untrusted sources.

Installing the WebUI and placing the checkpoint

  1. Use Git to clone the AUTOMATIC1111 WebUI repository to a folder on your PC.
  2. Download your chosen checkpoint file. These are large, often several gigabytes, so give it time.
  3. This is the make-or-break step. Place the checkpoint into the folder stable-diffusion-webui\models\Stable-diffusion. Nothing will appear in the model dropdown until a file lives here.
  4. Find webui-user.bat in the WebUI folder and double-click it. The first launch looks frozen because it is downloading several gigabytes of dependencies in the background, which can take anywhere from ten to forty-five minutes depending on your connection. Let it finish.
  5. When it is ready, the WebUI opens in your browser at a local address.

A 12GB-class card is the practical entry point for this. The GPU listings at Evetech make it easy to check the VRAM on any card you are considering against what local image generation actually needs.

Generating your first image

  1. In the WebUI, select your model from the checkpoint dropdown at the top. If it is empty, the file is not in the right folder; go back and check.
  2. Type a clear prompt into the text box. Be specific about subject, style and detail.
  3. Leave the size at 512x512 for your first run, since the base models are trained at that resolution and it generates fastest.
  4. Click Generate. On a 12GB RTX GPU the image appears in a handful of seconds.
  5. Iterate. Change the prompt, adjust the sampling steps, and regenerate. The fast turnaround on a local card is exactly why running it on your own machine beats waiting in online queues.

For anyone weighing whether their machine is ready, the AI PC builds at Evetech illustrate the kind of GPU and memory combination that makes local generation smooth rather than sluggish.

Choosing the right sampler and settings

AUTOMATIC1111 defaults to the Euler a sampler and 20 steps, which is a good starting point. Euler a produces varied outputs across runs, which helps when you are exploring a new prompt. For SDXL models, bump the resolution to 1024x1024 since those models are trained at that native size and produce sharper results there. The CFG scale controls how closely the image follows the prompt; a value between 7 and 9 is the typical range for most base models. Change one variable at a time so you understand what each setting does rather than adjusting everything at once.

Frequently Asked Questions

Why is my model dropdown empty?

The checkpoint file is not in the right folder. It must sit in stable-diffusion-webui\models\Stable-diffusion. Move it there and either restart the WebUI or click the refresh icon next to the dropdown.

How much VRAM do I need?

Around 12GB is a comfortable starting point for 512x512 generation with headroom. You can run on less, but you will hit limits faster on larger images or heavier models.

Why does the first launch look frozen?

It is downloading several gigabytes of dependencies in the background on first run. This is normal and can take ten to forty-five minutes depending on your internet speed. Let it complete rather than closing it.

Should I use safetensors or ckpt files?

Use safetensors. Older ckpt files from untrusted sources can carry malware, whereas safetensors is the safer modern format. Stick to reputable model sources either way.

How long does generating an image take?

On a 12GB RTX GPU, a 512x512 image takes only a few seconds once the model is loaded. Larger resolutions and more sampling steps increase the time, but a first basic image is quick.

Local image generation rewards a GPU with enough VRAM to keep things fast. Check your options in the AI PC range at Evetech to build a machine that handles Stable Diffusion comfortably.