Consistency, Not Talent

Why it matters

A good case in the wrong place still fails: before you build, you need the one-line claim that greets a visitor and a map of what goes where. And a consistent look is what separates a portfolio that feels intentional from one that feels thrown together; it comes from a handful of decisions, not from talent. The same goes for images: AI lets you make any image in seconds, which is exactly why judgment matters more than generation. Choose what serves your proof, reject the rest, and know when a real screenshot beats anything generated.

Brief

1. Write your one-line claim: the single sentence a visitor should remember. Use AI for ten options, then pick and sharpen one; the choosing is yours.
2. Build a content map: for each page, the sections in order, which case sits where (lead with your strongest), and the named call to action, all laddering up to the one action from Week 1. Note any proof you still need to gather (screenshots, a demo link, the repo, numbers, a testimonial).
3. Choose your type: one or two free fonts (a heading font and a body font; they can be the same).
4. Choose your palette: a main color, near-black text, near-white background, and at most one accent, calm enough that your work is the loudest thing on the page.
5. Make a simple logo or favicon, and write a two-line style note (fonts, hex codes, one sentence on the mood). Add it to your Claude Project so the build stays consistent.
6. List the images your portfolio actually needs, matching the content map. For your work, use real captures: clean, cropped, legible. For the connective tissue, generate options in one consistent style; for anything that is you, use a real photo. Curate ruthlessly and write one or two lines on a generated image you rejected and why.

Deliverable

The one-line claim, the content map (pages, sections, cases, calls to action) with the "still need to gather" list, the identity kit on one page (fonts named, palette with hex codes, the logo or favicon, the style note), and the final image set with your rejection note. Post it all in your track thread.

Evaluation criteria (pass/revise)

• The claim is single and memorable, not a paragraph.
• Every page has ordered sections and a named call to action, laddering up to the one action, and the gather-list is honest so the build week is not blocked.
• One or two fonts and a tight palette (about 3 to 4 colors) with actual hex codes.
• A simple logo or favicon exists, and the style note describes one coherent mood that frames the work rather than competing with it.
• The work is shown with real captures, not AI stand-ins, and any AI-generated images share one consistent style.
• The rejection note shows genuine judgment, not just "I liked this one."

Decide once, then judge ruthlessly

Use AI to give you options and to check your choices, not to decide for you. Generate a few image options, then reject most of them on purpose. That rejection is the actual skill.

Copy a prompt, fill in the {highlighted} parts, and remember: it gives options, you make the call.

Iris keeps it quiet on purpose

Iris wants her work to read as careful and clear, so she resists the urge to make the site itself flashy.

Iris rejected nine ideas to keep one. Her site ends up calm, and her museum map is the most colorful, memorable thing on it. Exactly right.

Dorian, on the marketing track, made the same call for a different reason: he wanted his before-and-after numbers to be the brightest thing on the page, so he picked one near-black, one warm white, and a single green accent for the "after" figures. The restraint is the point on every track.

Amateur, then intentional

Linked resources