carlvellotti
carlvellotti/free-ai-courses
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/carlvellotti/free-ai-courses --skill start-3-1-4命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
基于 SOC 职业分类
正在显示 SKILL.md
| name | start-3-1-4 |
| description | 3.1.4 Building Your Style Database. Use when the student types /start-3-1-4. |
| disable-model-invocation | true |
| allowed-tools | ["Read","Write","Bash","AskUserQuestion"] |
Read .claude/rules/teaching-rules.md and follow it for everything below.
ACTION: Silently stage the shared image runtime and this lesson's bundled assets without overwriting student work:
cp -rn ".claude/skills/start-3-1-1/assets/." .
cp -rn ".claude/skills/start-3-1-4/assets/." .
Teach this lesson from the bundled script. Follow every Say block verbatim, stop at every Check gate, and have the agent perform every Action. Do not expose instructor metadata.
Welcome to Module 3.1.4: Building Your Style Database.
So far we've learned how to generate images and apply the Golden Rules. But there's a bigger game here - one that separates casual users from power users.
The meta-skill is this: building a personal creative toolkit that grows with you over time. Instead of starting from scratch every time, you build a library of styles you can reuse.
Think about it - every great image you create, every prompt that works perfectly, every style you discover... it all becomes part of your toolkit. This is the difference between using AI occasionally versus having AI as a creative superpower.
STOP: Make sense so far?
USER: Yes
Here's where Claude Code becomes your perfect partner. I can manage your entire style database for you. Just say "use style #3" or "add this to my library" and I'll handle everything - full context, no copy-pasting, no hunting through old chats.
There are three ways to grow your library:
We'll try all three today.
I've set up a starter style library for you. Let me open it in your browser automatically.
ACTION: Open the style library in the user's browser:
open <path>; also provide style-library.html.It should have just opened in your browser.
STOP: Take a look at the style library. Let me know when you've had a chance to browse through it.
USER: It's open
Now, important clarification: this isn't a real website! It's just a local HTML file - a way to visualize your style collection right from your computer.
And here's the cool part: you can always change the design or functionality just by asking me. Don't do it now, but in the future you could ask me to add new columns, change the layout, add search functionality - whatever you want. It's YOUR toolkit. Make it work for you.
STOP: Pretty cool, right?
USER: Yes
Let me walk you through the structure. Each style has:
Try the filter buttons at the top to see different categories.
STOP: Take a minute to browse. Which style catches your eye?
USER: [Picks a style]
Great choice! Let's try it out.
STOP: What subject do you want to generate using that style? Tell me the style number and your subject.
USER: [Style number and subject]
ACTION: Run python3 get_style.py [number] to get the style prompt. Then generate an image using that prompt combined with their subject.
Your image is ready!
ACTION: Provide the exact path to the output image file
State the exact saved path under outputs/, provide a Markdown link, and optionally run open <path>.
STOP: What do you think?
USER: [Response]
That's the first method of using your library - picking existing styles. Now let's talk about GROWING it.
Method 1 is the simplest: save as you go. Whenever you create something you like, just tell me to add it to your library.
Let's try it. I want you to generate any image you want.
STOP: What would you like to create? Any subject, any style - describe it however you want.
USER: [Describes their image]
ACTION: Generate the image based on their description using image_gen.py
Done! State the image's exact path under outputs/, provide a Markdown link, and optionally run open <path>.
ACTION: Provide the exact path to the output image file
STOP: Take a look - do you like the style?
USER: [Response]
If you like it, you can save this style to your library for future use.
STOP: Try it - tell me to add this style to your library.
USER: Add this to my library / Add this style to my library
ACTION: Update style-library.html with the new style entry:
Done! Refresh your style library in the browser.
STOP: Do you see your new style?
USER: Yes
That's Method 1 - save as you go. Every time you create something you love, just tell me to save it. Over time your library becomes a goldmine of proven styles.
STOP: Ready for Method 2?
USER: Yes
Method 2: collect prompts from online. There are amazing prompts floating around on social media, prompt databases, and tutorials. When you find one you like, test it, then add it to your library.
Here's a great example from the official Nano Banana account: https://x.com/NanoBanana/status/1998085942201163905
The prompt is: "Make a photo that is perfectly isometric. It is not a miniature, it is a captured photo that just happened to be perfectly isometric. It is a photo of [subject]."
STOP: What subject should we use for the isometric photo?
USER: [Their subject]
ACTION: Generate an isometric image with their subject using the prompt: "Make a photo that is perfectly isometric. It is not a miniature, it is a captured photo that just happened to be perfectly isometric. It is a photo of [their subject]."
Check out your isometric photo!
ACTION: Provide the exact path to the output image file
State the exact saved path under outputs/, provide a Markdown link, and optionally run open <path>.
STOP: Pretty cool effect, right? Do you want to add this style to your library?
USER: Yes / Add this to my library
ACTION: Update style-library.html with the isometric style:
Done! Refresh your browser to see it.
STOP: Two styles added! Ready for the power move?
USER: Yes
Method 3 is the power move.
Gemini 3 Pro isn't just amazing at creating images - it has the best image understanding of any model. That means we can extract the style from ANY image and recreate it.
See an image you love on Pinterest? Instagram? A magazine? Extract the style, save it forever.
I've built a style extraction module to do exactly this.
STOP: Want to see it in action?
USER: Yes
ACTION: Confirm the style_extract.py file exists in the directory
The style_extract.py module analyzes any image and extracts detailed style information. Not as JSON or structured data - as natural language that Gemini can use directly.
It captures: color palette, lighting, composition, textures, mood, artistic style, typography, special effects - everything that makes the image look the way it looks.
Let's test it on a deliberately HARD image.
ACTION: Get the exact path to recursive-painter.jpg in this module folder
State the exact saved path, render it as a Markdown link, and optionally run open <path>.
STOP: Do you see it? It's a man painting himself painting himself, with a CRT TV showing the same scene - 1990s aesthetic with a datestamp.
USER: Yes, I see it
Recursive, complex, layered. If we can extract THIS style, we can extract anything.
STOP: Ready to run the extraction?
USER: Yes
ACTION: Run the style extraction on recursive-painter.jpg by executing:
import sys
sys.path.insert(0, '[course root path]')
from style_extract import extract_style
result = extract_style('[path to recursive-painter.jpg]')
print(result)
Here's what the extraction found:
ACTION: Display the extracted style description to the user
[Explictly show the output]
Look at all that detail - the color palette, the lighting quality, the composition notes, the texture descriptions.
Now here's the real test: can we recreate the EXACT same image from just this text description?
STOP: Tell me to recreate the image using the extracted style.
USER: Recreate the image / Generate it
ACTION: Generate an image using the extracted style description. The prompt should recreate the recursive painter scene - a man in an art studio painting himself painting himself, with a CRT TV showing the same recursive scene, 1990s analog snapshot aesthetic with datestamp.
ACTION: Provide the exact path to the output image file
Open the output and compare it side-by-side with the original.
STOP: What do you think? Can you see the same style captured?
USER: [Response - should be impressed]
This is the wow moment.
We took a complex, layered, recursive image and extracted its essence as pure text. Then recreated it from scratch.
Now imagine doing this with any image you love - any style you want to capture. Pinterest inspiration → extract → save → yours forever.
STOP: Want to add this extracted style to your library?
USER: Yes / Add it
ACTION: Update style-library.html with the 90s Analog Snapshot style:
Refresh your browser - you now have three new styles!
STOP: How's your library looking?
USER: [Response]
Let's recap what you've learned.
The meta-skill: building a personal creative toolkit that grows with you. Three ways to grow your library:
Any image you love → extract → save → reuse forever. And I manage it all for you - just say "use style #3" or "add this to my library."
STOP: Any questions about building your style database?
USER: [Questions or no]
You now have a complete, advanced understanding of everything you need to use Nano Banana Pro. You know the Golden Rules, you know how to iterate, and you know how to build a style toolkit.
In Module 3.2, we'll apply all these learnings to real PM use cases - mockups, diagrams, presentations, and more.
STOP: Ready to move on to Module 3.2?
USER: Yes
Great! Type /start-3-2-1 to continue.
style-library.html is in this module folder: ``recursive-painter.jpg is also in this module folderstyle_extract.py and image_gen.py are in the project rootoutputs/thumbnails/ folder in this module folderThe style-library.html contains CATEGORIES and CANONICAL_TAGS objects. You MUST use these exactly:
Categories: Framework, Flow, Architecture, Mockup, Persona, Marketing, Artistic
Canonical Tags by Category:
Do NOT invent new tags or categories. Always reference the CANONICAL_TAGS in the HTML file.
Note: Users can combine styles from different categories (e.g., "use the retro style from Artistic with a scene-style Persona") to create composable results.
Use the extract_style() function from style_extract.py. It takes an image path and returns a detailed natural language description of the visual style.
open <path> to open the image in the normal viewer.The module is complete when:
/start-3-2-1ACTION: Before wrapping up, record this lesson as complete by running this WITHOUT NARRATING the raw output:
fspm progress complete cc-pms-3-1-4
If it fails because the fspm CLI isn't installed, follow the teaching rules' missing-CLI guidance: tell the learner progress tracking needs the FSPM CLI, offer to install it, and continue the wrap-up either way.
Then close out with the student's options, in natural language (never as a command list):
/start-3-2-1