Stable Diffusion online
Stable Diffusion is an open image model that turns a written prompt into a picture. Stability AI released it in August 2022 with researchers from the CompVis group at LMU Munich and Runway. It was the first high-quality image generator anyone could download and run on their own computer, and that openness grew one of the largest creative communities in AI.
Hotpot no longer runs Stable Diffusion. This page explains how Stable Diffusion works, the free ways to use it, and what its settings mean. If you only want to make an image from an idea, the AI Image Generator works in your browser with nothing to install.
How Stable Diffusion makes an image
Every image starts as random noise. The model removes a little of that noise at a time, and at each step your prompt nudges the result toward what you described. After enough steps a picture emerges from the static.
The trick that made Stable Diffusion practical is where this happens. Instead of working on full-size pixels, it works on a compressed version of the image called a latent, which is many times smaller. A decoder then expands the finished latent into the final picture. That is why it is called latent diffusion, and why it fits on a home graphics card.
The starting noise comes from a number called the seed. With the same seed, prompt, and settings, Stable Diffusion draws the same image again. Change only the seed and you get a new variation of the same idea.
Stable Diffusion versions
Each release changed what the model is good at. These are the ones you will see mentioned most.
- Stable Diffusion 1.5 (2022) is small, fast, and still popular because thousands of community models are built on it.
- Stable Diffusion 2.1 (2022) raised the native size to 768 pixels and changed how prompts are read, so many 1.5 prompts behave differently.
- SDXL (2023) moved to 1024 pixels with better composition, lighting, and hands. It became the new base for community models.
- SDXL Turbo (2023) makes an image in as little as one step, fast enough to update while you type.
- Stable Diffusion 3 and 3.5 (2024) use a new architecture that follows long prompts more closely and writes short text inside images more reliably.
Free ways to use Stable Diffusion
On your own computer
Running it locally is free, private, and unlimited. You need a recent graphics card with several gigabytes of video memory, and larger versions need more. Apple Silicon Macs can run it too, more slowly. Popular free interfaces include ComfyUI, which builds images from connected blocks, and AUTOMATIC1111’s Stable Diffusion web UI and its offshoot Forge, which use classic forms and sliders. You download the model files separately, usually from Hugging Face.
On a rented GPU
Hosted notebooks and cloud GPU services let you run the same interfaces on someone else’s hardware by the hour. This suits people without a strong graphics card who still want full control.
On a website
Many sites run Stable Diffusion for you. Graphics cards are expensive to run, so free sites usually pay for it somehow. Expect a daily limit, a queue, ads, smaller images, or an older version. Before relying on one, check which version it runs and what its terms allow.
Stable Diffusion settings, explained
Most Stable Diffusion tools expose the same handful of controls.
- Prompt describes what you want to see. Early versions rewarded lists of keywords. Newer versions understand full sentences better.
- Negative prompt lists things to avoid, such as blur, extra fingers, or text.
- Steps set how many rounds of noise removal run. More steps add detail up to a point, then mostly add time.
- CFG scale, also called guidance, sets how strictly the image follows the prompt. Low values drift and get creative. High values follow closely but can look harsh and oversaturated.
- Seed fixes the starting noise so you can repeat or vary an image on purpose.
- Sampler is the method used to remove noise. Different samplers trade speed for detail, and some look good in fewer steps.
- Image to image starts from your picture instead of pure noise. Denoising strength sets how far the result may stray from it.
- Inpainting regenerates only a masked area, for fixing a hand or replacing an object.
- LoRA files are small add-ons that teach the model a face, character, object, or style without retraining the whole thing.
- ControlNet guides the layout with a pose, sketch, depth map, or outline.
- Upscaling enlarges a finished image and adds detail, since most versions create images around 512 to 1024 pixels wide.
Making images without the setup
Stable Diffusion rewards people who enjoy tuning. If you would rather describe an idea and get a picture, Hotpot handles the technical side. There are no installs, graphics cards, samplers, or step counts, and it works on a phone as well as a computer. Your first image does not need an account.
If you know Stable Diffusion, here is where the familiar pieces live on Hotpot.
| In Stable Diffusion | On Hotpot |
|---|---|
| Prompt and negative prompt | Write one plain description in the AI Image Generator, including what to leave out, such as “no text” or “plain background” |
| Checkpoints and style LoRAs | Pick from hundreds of image styles and art styles, or add your own style in the Library |
| Character LoRAs | The AI Character Generator and Fantasy Character Generator |
| Image to image | AI Style Transfer keeps your picture and changes its look |
| Inpainting | The AI Image Editor changes part of an image from a written instruction, and the AI Eraser removes objects |
| Upscaling | The Photo Upscaler |
Prompt habits worth keeping
Long strings like “masterpiece, best quality, 8k, ultra detailed” helped early Stable Diffusion models, which were trained on short captions and tags. Newer image models respond better to a clear sentence. Name the subject, the setting, the light, and the medium, in that order.
A red fox curled up in fresh snow at dawn, soft pink light, close-up wildlife photograph, shallow depth of field.
Change one thing at a time when refining. Swap the light, then the framing, then the style. You will learn faster what each word does.
Questions
Is Stable Diffusion free?
The model files are free to download. Older versions use an open license, and newer ones use Stability AI’s community license, which is free for individuals and smaller businesses. Read the license for the version you choose before using images commercially.
Can I use Stable Diffusion online without logging in?
Some sites allow a few images without an account, usually with limits. Running it on your own computer needs no account at all.
Does Hotpot use Stable Diffusion?
No. Hotpot no longer runs Stable Diffusion. Its image tools are built differently and need no settings or setup.
What computer do I need to run Stable Diffusion?
A desktop or laptop with a recent graphics card and several gigabytes of video memory. Smaller versions run on modest cards, and Apple Silicon Macs work more slowly.
Start with an idea
Create photos and images or explore paintings and illustrations. For more on writing prompts, read how to create AI images and art.