INTD 318W07SETS THE 20%

Art
direction.

AND WHOSE DEFAULTS YOU ARE SHIPPINGMON 2 NOV 2026

Prep: volunteered venture agreed at the end of W06; its visual context built in advance from real material; both demo rounds rehearsed; contrast checker open; handout printed; lens cards; DESIGN-SYSTEM.md open in a window; the W08 brief printed and ready to hand out.

Second CM-owned session, with no precedent in Ryan’s corpus. It sets the largest single deliverable in the course, due in two weeks. Everything today points at that.

Prototypes on screens.

ONE SENTENCE EACH
What W06 testing said, and what you changed because of it. Not what you plan to change.
SCREENS OPEN, NOT DESCRIBED
The thing itself, on the screen. This is the last session where anyone describes their prototype instead of showing it.
NOTHING BUILT YET?
Then you art-direct what you are about to build. Today works either way. The context comes first regardless.

Keep the fallback visible and unembarrassing. W06 is Ryan’s, and the prototyping-tool decision has been open all term. A student with nothing on screen is not necessarily behind, and today does not depend on it.

Look at what is on the screens as they go. Expect the generic default in eight of eighteen ventures. Naming one of them during the demo is worth more than any example you could bring in.

It works.
And it looks like
nothing in particular. →

THAT IS TODAY’S PROBLEM, AND IT IS NOT A TASTE PROBLEM

The bridge out of the standup. Say it without contempt. What they built works, which was the ask in W06. The gap is that it carries no position, and a venture whose premise is that it differs from what exists cannot afford to look like everything that exists.

Flag the change of register early. This is the first session where the subject is the artifact rather than the evidence behind it. It is still the same argument.

INPUT · 15 MIN

Every visual decision
is a claim.

Fifteen minutes. Instructor talk today totals forty-five minutes across three blocks, at the fifty-minute cap. The studio is ninety minutes because of that.

A typeface is a statement about attention.

HOW MUCH, AND FROM WHOM
How much attention you are asking for, and who you are asking it of. That is a claim about your customer, made before they read a word.
THREE REAL PRODUCTS
Not “it’s nice.” What is it asserting? Say the claim out loud for each one.
HIERARCHY IS A DECISION
What someone sees first is something you decided, whether or not you decided it deliberately.

Have the three products chosen and the sentence written for each. Doing this live is how fifteen minutes becomes twenty-five.

Pick at least one that is deliberately loud and one that is deliberately quiet, so “good” is visibly not a single direction.

Colour carries meaning before it carries mood.

AND THE MEANING IS CONTEXTUAL
The same green reads as safe in a banking product and cheap in a pharmacy one. No palette is good in the abstract.
SO “PICK NICE COLOURS” IS NOT A BRIEF
Named roles are a brief: ink, ground, one accent, and the states that accent has to carry.
ROLES SURVIVE, PALETTES GET REPLACED
You will change the hex values. If the roles are named, that costs an afternoon. If they are not, it costs a rebuild.

This is the row students skip in the deliverable, so it gets a slide. A mood board is not a palette. A palette says what each colour is for.

Accessibility is part of the design.

NOT A CONSTRAINT ON IT
Contrast. Minimum sizes. Never signalling state by colour alone.
TEN SECONDS, AND A RIGHT ANSWER
Watch. Check a ratio live, on screen, now. This is the one part of today’s session where you can be provably right.
WHICH IS WHY IT IS A GATE
Contrast checked and the figures recorded is a pass condition on the W08 hand-in. Not a criterion. A gate.
Same type. Three grounds. Ten seconds to check.Start your ventureBody copy at eighteen pixels.2.1:1FAILSLooks fine on a bright laptop. Unreadable on a phone outdoors.Start your ventureBody copy at eighteen pixels.5.9:1PASSES AAThe floor for body text at normal sizes.Start your ventureBody copy at eighteen pixels.14.8:1PASSES AAAThe deck you are reading right now.

Run the checker on screen. The point is not the number. It is that the check is cheap. Most students assume accessibility work is a project, and ten seconds of demonstration ends that assumption.

Use a palette that looks fine on the projector and fails. That is the honest case, and it is common in generated palettes.

SET-PIECE · 15 MIN

Watch this.

The fifteen minutes that has to land. Two rounds, one venture, same request. Rehearse beforehand on the model and the tool you will use.

Know the two failure modes before you start. They are in the notes on the round-one slides.

Round one. What a student would type.

Design a clean, modern landing page for [the volunteered venture]. Make it look premium and trustworthy.

It will produce something. It will be competent. Say so out loud.

You asked for“clean, modern, premium”It decidedTypefacePalettePhotographyTone of voiceLayout gridButton shapeImagery styleDensityIconographyMotionCopy registerWho it is forTwo words in. Twelve decisions out. None of them yours.You have evidence for every one of them. You just never wrote it down where it could see it.

Say it is competent, and mean it. The credibility of the next ten minutes depends on not pretending the output is worse than it is. A room of designers can see the screen.

This is where 317’s version of the demo does not transfer. 317 could rely on a thin cold prompt producing something visibly poor. A cold visual prompt in 2026 does not. The lesson has to survive good-looking output, and it does. That is the next slide.

Two questions. In this order.

1WHAT DID IT DECIDE THAT WE DIDN’T TELL IT TO?The palette. The typeface. The photography. The whole implied customer. Count them out loud. There are usually a dozen.
2WHOSE PRODUCT DOES THIS LOOK LIKE?Someone will name a real company within five seconds. That is the finding.

Question two carries the weight, and it works better on polished output than on poor output. That is what makes the demo robust. If round one came back surprisingly good, lean on question two.

If round one came back poor, do not celebrate. That is the cheap version and it teaches the wrong lesson. Note it, move on, and carry the weight on the twelve-decisions count instead.

Round two. Same request, with the context.

Goal, stated as an effect on a named person · three references marked good and three marked bad, with a reason each · the constraints · the non-goal. Then: the same request.

Put the two side by side and leave them there for the rest of the session.

The context for this venture was built before class from their real W02 to W05 material. Show it briefly as a document. The students are about to write one in thirty minutes and need to have seen the shape.

Do not narrate the comparison while it renders. Let them look.

It didn’t get smarter.
It stopped guessing. →

YOU HAD THE EVIDENCE. IT JUST COULDN’T SEE IT.

The full line: “The second one isn’t better because the model got smarter. It’s better because it stopped guessing. Every one of those twelve decisions it made for you in round one is a decision you have evidence for. You did the interviews. You never wrote it down where the tool could see it.”

Not an anti-AI moment, and not a prompt-tips session. The studio block is the payoff: the same tool, their own evidence, work they could not do as fast alone.

This session’s instance of the frame.

THE LIFT

Twenty directions in the time it takes to build one, which gets you past your own first instinct, reliably the most conventional one you have. And applying a decision across fifteen screens without getting bored, which is where people actually fail.

THE LIE

It returns taste-shaped output, the visual average of its training data, and presents it as a direction. It will hand you palettes that fail contrast while looking fine on a bright laptop. And it cannot see its own output.

THE CATCH

You have five weeks of evidence about who this is for and what they need to feel. It has none of it. And you can measure contrast in ten seconds.

“It cannot see its own output” is the sharpest line on this slide for a visual session. The model describes what it intended, not what it made. Say it slowly.

Same boundary as W03: the frame applies to model output, not to evidence students gathered themselves. It stays separate from Ryan’s 4-Risks. Different axis, not a level below.

Six parts.
Most people supply one. →

A direct port of 317’s anatomy of a working context, moved into the visual register. The handout they keep lists the same six.

The visual context.

GOAL
What this has to do, to a named person, in a real setting. Not “look premium” but “read as trustworthy to a 55-year-old checking it on a phone in a waiting room.”
CONSTRAINTS
Surface, medium, the accessibility floor, and anything already fixed: a name, a platform, a partner’s brand sitting next to yours.
REFERENCES
Three good, three bad, one sentence each on what specifically. The bad ones do more work and everyone skips them. “Not like this” is more legible than any adjective.
NON-GOALS
The aesthetic you are deliberately not chasing, and who you are content to lose.
HOUSE RULES
The corrections you have had to make twice. Starts empty. Fills up over the next fortnight.
STATE
What exists right now, and what is still a sketch.

Spend the time on REFERENCES. It has the highest return and it is the part everyone does badly. Three bad references with a reason each will change their output more than anything else on this slide.

The GOAL example is deliberately over-specified. Read it as written. The contrast with “look premium” is the point.

If you correct the same
thing twice, that is a
context bug. →

FIX IT UPSTREAM AND IT STOPS HAPPENING

317’s line, unchanged. It is the difference between someone who prompts and someone who builds a system.

Connect it forward explicitly: a correction you make twice is a house rule, and house rules are what the W08 deliverable asks for. This sentence makes the assignment make sense.

A design system is three to five sentences.

Open the real one. Not a slide about it. The document. Its three rules, and the fact that each one is a sentence you could point at a screen and check.

Orange means the course is asserting something · the title never moves · a box carries a fill, or a stroke, and never both.

The most credible five minutes in the session. A real system, in use, for the course they are sitting in, small enough to read on one screen.

The teaching point: each rule removes a category of decision permanently. That is what makes a system worth having at eighteen-student, five-week scale. Not documentation. Fewer decisions.

BREAK · 10 MIN

Back at 4:35.

Hand out the visual-context sheet and the W08 brief now, on the way out. They read them in the break and the studio starts faster.

STUDIO · 30 MIN

Build yours.

Six parts. Your venture. Your own material.

FROM W02 TO W05, NOT FROM SCRATCH
You already did the interviews. This is transcription of what you know into a form the tool can read. It is not new research.
THE REFERENCES ARE THE WORK
Three good, three bad, a sentence each. If you finish early, you did the bad ones badly.
GOAL IN ONE SENTENCE, NAMING A PERSON
If your goal has an adjective and no person in it, it is not finished.

Circulate on the goal sentences first. That is where the failures are, and a bad goal makes the next fifteen minutes worthless. Look for adjectives standing in for people.

Second pass: the non-goal. Almost everyone leaves it blank. “Who are you content to lose?” is the question that unblocks it.

Now run it.
Twice. →

Generate against it. Then generate cold.

WITH THE CONTEXT
Paste it in. Make the same request you would have made anyway.
THEN WITHOUT IT
Clean thread. “Design a landing page for [your venture]. Clean and modern.”
KEEP BOTH
Both go on the wall in twenty minutes, and both go in the process book. This is the comparison the Evidence Handling row asks about.

Order matters. Context first. Reversed, they anchor on the cold output and read the directed one as a variation of it.

Some will find their cold output fine. Send them to question two: whose product does it look like? That is the finding, and it lands harder on their own venture than on the demo.

PANEL · 25 MIN

Four lenses.
Rotate.

Groups of four. Every screen gets all four lenses, and the lenses rotate each round. Lens cards on the tables before they sit down.

The technique is 317’s named-lens review panel. Vague critique produces vague feedback. An assigned lens makes people find things.

Assigned. Not “what do you think.”

CRAFT

Is this well made, or just confidently arranged?

FIT

Does this look like it belongs to this venture, or to any venture?

ACCESS

Who cannot read this, and why? Contrast, size, and anything signalled by colour alone. Measure it.

TRUST

Would you give this your credit card?

TRUST is the lens that makes this a 318 session rather than a borrowed 317 one. It is where the design axis and the business axis land on the same screen. Students can answer it honestly about each other’s work in a way they cannot answer “is it good.”

ACCESS is measurable. Insist on a number, not an impression. It is the one lens with a right answer, and that makes the whole panel less polite.

CLOSE · 30 MIN

Side by side.
On the wall.

Directed, and undirected.

BOTH UP, EVERY VENTURE
Two minutes each. Not a presentation. One sentence on what the context changed, and one on what it did not.
“THE COLD ONE IS FINE” IS AN ANSWER
Then say whose product it looks like. If nobody can name one, that is worth the room’s attention.
WATCH FOR THE HOUSE RULES
Anything you corrected twice this afternoon is a rule. Write it down before you leave. You need three to five of them by the 16th.

Eighteen students at two minutes each is thirty-six minutes, which is too long. Run it in threes on the wall as in W03, or take six volunteers and let the rest pin up without presenting. Decide before the session.

Due the 16th. The whole floor.

COLOUR
Named roles. Contrast checked, figures recorded.
TYPE
Two families maximum. Four or five sizes, jumps stated.
SPACE
One unit, and the multiples in use.
THREE COMPONENTS
In their real states. The three that carry your prototype.
THREE TO FIVE RULES
Sentences. Falsifiable. This is the row people skip, and it is the one being weighted.
APPLIED
The prototype, visibly built from all of the above. No orphan screens.
Colour: named roles, not a mood boardInkGroundAccentMutedType: four sizesAa34 pxAa26 pxAa20 pxAa16 pxSpace: one unit1 × 8px2 × 8px4 × 8pxThree components, in their real statesContinueContinueContinueEmail addressNeeds an @Card titleOne line of supporting copy.And three to five rules. The row people skip1 · Accent means the product is asserting something. Never decoration.2 · A box carries a fill, or a stroke, and never both.3 · Nothing is signalled by colour alone.That is the whole floor. Everything above fits on one page, and the rules are what make the next decision easy.

20% of the course. The brief and the rubric are both on the course site. Point at them explicitly, and say that the rubric is the specification.

Say the gates out loud: AI Use Statement, the visual context submitted alongside, contrast figures recorded, and a prototype link that opens. A missing gate returns the work unmarked, not marked down.

Write down the thing
you fixed twice. →

W08 · MON 16 NOVDESIGN BUILD · SYSTEM DUE

The handoff line, at 6:10 and not 6:14: “Two weeks. The context you built today is what you generate against, and the system you hand in on the 16th is what that context hardened into. If you correct the same thing twice between now and then, that correction is a rule. Write it down. That is the whole assignment.”

W08 is also a CM session, so this seam is internal. It carries the graded artifact, and it should not be delivered over the sound of eighteen people packing up.

Speaker notes

next
back
scroll next & back
swipe next & back
s speaker notes
o overview & jump
t start the 3:25 clock
f fullscreen
esc close

Turn your phone sideways

These slides are a fixed 16:9 canvas. In portrait they shrink to an unreadable strip; landscape gives you the whole slide.

Sessions