L09 | English F3 Student Learning Log
L09 | English F3 Student Learning Log
Persona: I'm a Form 3 (Grade 9) student and I only read English. I know basic Python (variables, for loops, lists) and school algebra up to quadratics. I learn from the official pages, the videos with captions, and the book. Kaggle notebook steps are written as a practice log that follows the official notebooks.
What I understood (in my own words)
- Stable Diffusion turns a sentence into an image.
- It starts from pure noise and removes a little noise at each step until a picture appears.
- CLIP teaches the model how words and images match up.
- Part 2 is hard. I get the overall flow, but for the maths I can only learn the names for now.
Where I got stuck, and how I got unstuck
| Where I got stuck | How I solved it |
|---|---|
| Part 2 maths is way beyond school level | I started with each lesson's 'Concepts discussed' list, looked up each term, then watched the first half of the video. |
| What are latents? | I think of them as a compressed code for the image: much smaller, so faster to work on, then decoded back into a picture. |
My answers to the exercises
Q1. Name Stable Diffusion's four core components and describe each in one sentence.
- CLIP text encoder: turns the prompt into numbers. 2) VAE: compresses images into small latents and decodes them at the end. 3) U-Net: predicts which part of the current image is noise. 4) Scheduler: decides how much noise to remove at each step.
Q2. How do generated images change when the guidance scale is high vs low?
High: the image follows the prompt closely but may look over-saturated or unnatural. Low: more varied and creative, but may drift from the prompt.
Q3. You want 'a dog on a beach' with no people. Write the prompt and negative prompt.
Prompt: 'a happy golden retriever running on a sunny beach, photo'. Negative prompt: 'people, person, crowd'.
My Zettelkasten card
Card｜Generating an image is really erasing noise, step by step
Stable Diffusion doesn't paint the whole image at once. It starts from noise, asks the model 'which part is noise?', removes a bit, and repeats dozens of times until a picture appears. What looks like magic is many small corrections.
It's like sketching in art class. Our teacher told us to draw lots of light, loose lines first, then erase what we didn't need and darken what we did, until a face emerged. When I tried to get it right in one stroke, the proportions were always off. I'll be honest: the Part 2 maths, like finite differences and analytic derivatives, is beyond me right now. I'm noting that both are about 'how to change the image to make it better' and leaving the details for later.
My next step: generate the same prompt with guidance scale 3, 7.5 and 15 and compare the three images side by side.
🔗 Related notes to grow next: → Art: 'Good work comes from repeated correction' → Ethics: 'Generated images mean seeing is no longer believing' → Learning: 'Map hard topics first, fill in details later'
Back to lesson note：L09_Stable_Diffusion入門_202611041526
