Learn Stable Diffusion in 5 minutes.
Stable Diffusion is the open image model you can run anywhere: locally, on a server, or through community interfaces like AUTOMATIC1111 and ComfyUI. The lesson drills the vocabulary that controls it: checkpoints, CFG scale, seeds, negative prompts, and img2img.
Ole is taking you through this one. Your docent for this room. Free, and it runs right here.
Lesson not loading? Open it in a new tab. It is the same free lesson either way.
Checkpoint, prompt, or dial?
With Stable Diffusion you own the whole stack, which means a bad image can be the model's fault, the prompt's fault, or a slider's fault. Diagnosing which is the skill this lesson builds.
Checkpoints and LoRAs
A checkpoint is the trained model file; community fine-tunes specialize in anime, photorealism, or design, and LoRAs bolt a style or subject onto any of them. Model choice sets the ceiling.
Guidance scale
CFG scale controls how strictly generation follows your prompt. Low values wander creatively; very high values fry the image. Most useful work happens in the middle of the range.
Start from an image
Instead of noise, generation starts from a picture you supply, with a denoising strength slider deciding how much of it survives. It is the bridge from a sketch or photo to a finished render.
What you will do in the lesson
You answer and try things. Nobody demonstrates at you for five minutes.
- Meet the moving parts. You map prompt, checkpoint, sampler, steps, and seed onto what each one actually decides, so settings stop being superstition.
- Hold the seed, change one thing. You fix the seed and vary a single setting to see its isolated effect, the experimental habit every Stable Diffusion user needs.
- Sweep the CFG scale. You watch one prompt at low, medium, and high guidance and learn to recognize an overcooked image on sight.
- Steer with a negative prompt. You strip watermarks, extra limbs, and mush with the negative field instead of rewriting the positive one.
- Transform an image with img2img. You pick a denoising strength that keeps the composition while changing the style.
Before you start
Do I need a powerful GPU?
Only for running it locally, and even then modest consumer cards handle smaller models. Plenty of hosted services run Stable Diffusion in the cloud, and the concepts in this lesson apply identically either way.
Do I need to install Stable Diffusion to take this lesson?
No. The lesson runs entirely in this page, free, with no setup. Installing a local interface afterwards is optional, and because the model weights are open there is a free way to run it yourself.
What is the difference between AUTOMATIC1111 and ComfyUI?
They are community interfaces to the same models. AUTOMATIC1111 is a settings-panel style web UI, while ComfyUI wires generation as a node graph you can rearrange. Beginners usually start with the former; pipeline builders graduate to the latter.
How does this compare to Midjourney?
Midjourney is a curated service with one evolving model and a short list of parameters. Stable Diffusion trades that polish for total control: your hardware, your checkpoints, your fine-tunes. It is the difference between renting and owning.
Other tools people learn next
Same format, same five minutes, still free.
Learners from these organizations are already learning on 
