How to use Stable Diffusion?
To use Stable Diffusion, pick an interface (a free web demo, or a local install like AUTOMATIC1111 or ComfyUI), load a model checkpoint, type a text prompt plus a negative prompt, set steps and CFG scale, then click generate. Beginners should start with a browser tool; power users install it locally for unlimited free images.
Why — the first-principles explanation
Using Stable Diffusion means feeding a text prompt into a model and adjusting a few dials that control the output. Every interface, from a website to a local app, exposes the same core controls, so learning them once transfers everywhere.
The first choice is where it runs. A hosted website requires no setup but may cost credits and limits customization. A local install (AUTOMATIC1111's WebUI or ComfyUI) is free, private, and unlimited, but needs a capable GPU. Either way, you then select a model checkpoint, which is the trained file that determines the art style, from photorealistic to anime.
The main dials are simple. The prompt describes what you want; the negative prompt lists what to avoid. Steps (around 20 to 30) set how many denoising passes run. CFG scale (around 5 to 9) sets how strictly the image follows your words. Sampler and seed control the algorithm and randomness; reusing a seed reproduces the same image.
Beyond text-to-image, most tools offer img2img (transform an existing picture), inpainting (repaint part of an image), and add-ons like LoRAs and ControlNet for style and pose control. Start simple with text-to-image, then layer in these features as you get comfortable.
An example that makes it click
It's like using a fancy coffee machine. The prompt is telling it 'a strong latte with oat milk'. The negative prompt is 'no sugar'. Steps and CFG scale are the strength and temperature knobs. You press the button, taste the result, and adjust the knobs a little next time.
Your first cup might be watery. You bump the strength up, change one word, and try again. After five or six tries you dial in a recipe you love, then you save that seed and settings so you can make the exact same cup whenever you want.
How to do it
- Choose an interface: a free web demo to start, or a local install (AUTOMATIC1111 or ComfyUI) for unlimited free use.
- Load a model checkpoint that matches your desired style (realistic, anime, art).
- Type a clear text prompt describing your subject, style, lighting, and composition.
- Add a negative prompt listing things to avoid, like 'blurry, extra fingers, watermark'.
- Set steps to about 20 to 30 and CFG scale to about 5 to 9.
- Click generate, then refine your prompt and settings and regenerate until satisfied.
- Use img2img or inpainting to edit results, and save the seed to reproduce a favorite image.
Key facts
- AUTOMATIC1111 WebUI and ComfyUI are the two most popular free local interfaces for Stable Diffusion.
- Typical starting settings are 20 to 30 sampling steps and a CFG scale of about 5 to 9.
- The seed value controls randomness; reusing the same seed and settings reproduces the same image.
- Model checkpoints determine the output style and can be swapped freely.
- Common modes include text-to-image, img2img, and inpainting for editing parts of an image.
The open-source image model you can run on your own hardware.
Affiliate link — we may earn a commission at no cost to you.▶ The 60-second explainer (script)
Here's how to use Stable Diffusion, step by step. First, decide where to run it. Beginners can use a free website with no setup. If you want unlimited free images, install a local interface like AUTOMATIC1111 or ComfyUI on a computer with a good graphics card. Next, load a model checkpoint, which sets the art style, realistic, anime, whatever you like. Now the main part: in the prompt box, describe what you want, the subject, the lighting, the style. In the negative prompt box, list what to avoid, like blurry or extra fingers. Set steps to around twenty-five and CFG scale to around seven, then hit generate. Look at the result, change a word or a setting, and generate again. That loop of tweak and regenerate is the whole skill. Once you get an image you love, save its seed so you can recreate it exactly. Later, explore img2img and inpainting to edit images instead of starting from scratch.
What authoritative sources say
People also ask
Do I need to code to use Stable Diffusion?
No. Interfaces like AUTOMATIC1111 and ComfyUI, and hosted websites, give you buttons and text boxes. No coding is required.
What settings should a beginner use?
Start with about 25 steps, CFG scale 7, and the default sampler. Change one thing at a time so you learn what each dial does.
How do I make the same image again?
Reuse the exact prompt, model, and seed value. The seed locks the randomness so the output is reproducible.
What's the difference between text-to-image and img2img?
Text-to-image builds a picture from words alone. Img2img takes an existing image and transforms it based on your prompt.