NCAD AI Workshop Week 03 / 15

All tracks · 4 hours · drop-in friendly

Research, Moodboards, and Concept Development Without Copying

Use AI to research wider and ideate faster — with verification habits that catch invented sources, and an influence method that keeps you on the right side of copying.

▶ HTML Slides

01This Week at a Glance

You will make: a mini concept board + verified source list + 150-word direction statement.

1 · Concepts (40 min)

Research hygiene + SIFT; the flattened-references problem; influence vs. copying; provenance moodboards.

2 · Demo (45 min)

"Make it cyberpunk" → three real territories — with a reference verified (or busted) live.

3 · Lab (70 min)

Influence Triangle hunt → concept board → direction statement.

4 · Share (45 min)

The narrate-the-path test.

Why this matters in real work

Research is where AI gives the biggest honest speed-up — and where it fails most dangerously, inventing sources with a straight face. One invented reference in a crit or client meeting destroys more credibility than ten good ones build. AI widens the search; you verify the sources and own the narrowing.

02Key Concepts

1 · Research hygiene: every claim is unverified until checked

The reliable failures: invented references (artists, books, movements that don't exist), plausible-wrong history, and citations that don't say what's claimed. The counter-move is SIFT — 60 seconds per claim:

  1. Stop — before repeating or reusing a claim.
  2. Investigate the source — who actually says this?
  3. Find better coverage — does anyone reputable agree?
  4. Trace to the original — click through to the actual image/text.
Watch out

"But it gave me a citation!" — cited ≠ correct. Tools cite pages that don't say what's claimed. Trace to the original; that's the T.

2 · The flattened-references problem

"Moodboard ideas for X" returns the same board to everyone — the average of the internet. Three escapes: ask for territories (named directions with qualities), demand the non-obvious ("avoid the clichés of [aesthetic]; flag uncertain references"), and cross domains — geology, deep-sea biology, industrial processes. The model has read those corpora too; almost nobody asks.

3 · Influence vs. copying: the working test

Legitimate influenceCopying
Multiple sources, transformed and combinedOne source, imitated
Qualities extracted ("the light," "the geometry")Signature lifted (someone's whole look)
Historical/public-domain + cross-domainA living artist's lane, occupied
You can narrate the pathYou hope nobody sees the reference
The influence triangle

One historical source + one contemporary source (engaged with, never imitated) + one non-art source → a direction none of them contains alone. Three points make a territory; one point makes a tracing. Today's lab.

4 · Moodboards with provenance

With AI in the mix, every board item gets a mark: [real — verified], [real — unverified], [AI-generated]. A synthetic "1970s Polish poster" that never existed will quietly art-direct your whole project from inside your moodboard.

Watch out

AI-generated "references" feel like research but are outputs — the average machine feeding you its averages dressed as history.

5 · The direction statement: the narrowing is yours

150 words: the direction, what it draws from (named, verified), what it refuses, why it fits the project. The model can describe territories; choosing and committing is the authorship act.

03This Week's Words

HallucinationIn research mode: fake artists, books, movements — confidently delivered.
Retrieval / RAGGrounding the model in documents you provide (NotebookLM = visible RAG). Shrinks hallucination; doesn't eliminate it.
Grounding / citationTying claims to checkable sources — then checking them.
Deep research modeCompiled, cited reports. Draft maps — every kept citation still gets clicked.
MoodboardCurated references defining a direction. Now needs provenance marks.
Visual territoryA named, specific direction with qualities, references, exclusions.
ProvenanceThe traceable origin of an asset or claim.
Influence triangleHistorical + contemporary + non-art → a new direction.
Direction statement150 words committing to a direction. The authorship act.
Public domainOut-of-copyright work — the deep, legitimate influence mine.

Full course glossary →

04Demo: "Make It Cyberpunk" → Three Real Territories

I made this, so can you

A weak aesthetic word becomes three named, verified territories — and one AI-suggested reference gets busted live.

Prepared demo artifact

Instructor backup / finished example: visual-territory-cards.html hallucinated-reference-example.md. Use it if the live tool run fails, or to reveal the finished structure after critique.

  1. The weak ask, exposed. "Cyberpunk venue identity, 5 prompt ideas" → neon, rain, kanji, hooded figure. The average of the internet, served instantly.
  2. The territory expansion. Three named territories for "near-future tech unease" — each with 5 concrete qualities, one historical reference, one contemporary practice, one non-art reference, an avoid list — plus "flag any reference you're not certain exists."
  3. The verification, live. Two references get image-searched on the projector. Typically one is real-but-different or a ghost. The card gets corrected with what the search actually found — the correction is the method.
What to notice

Specificity beat vibes; the non-art reference is where originality snuck in; the uncertainty flag changed the model's honesty; verification took 90 seconds and saved the direction.

05Step-by-Step Workflows

Verified research sprint (25 minutes, reusable forever)

  1. Map (5 min): lay of the land — traditions, 5 names, 3 debates, beginner mistakes, uncertainty flags.
  2. Question-find (3 min): "the 5 most useful questions for THIS project." AI is often better at questions than answers.
  3. Verify (10 min): SIFT what you intend to keep; image-search visual references; kill what fails.
  4. Source list (5 min): what survived — source, where verified, what quality you're taking.
  5. Hand off to taste (2 min): pick the 2–3 verified threads that excite you. The model doesn't get a vote.

The influence triangle

  1. Pick the feeling/problem of your project — not an aesthetic name.
  2. Historical: one verified pre-1970s (ideally public domain) source with a nameable quality.
  3. Contemporary: one current practice you engage with — what problem are they solving? Never a style to lift.
  4. Non-art: one source from science, industry, nature, sport, food.
  5. Synthesize: write what the three make together that none contains alone.

06Prompt Recipes

The territory expander"[weak aesthetic word]" is doing no work in my brief. Act as a design researcher. Give me THREE distinct visual territories that could express [the actual feeling/problem] for [project + audience], deliberately avoiding the clichés of [aesthetic word]. For each territory: a name, five concrete visual qualities (materials, light, colour behaviour, type feeling, texture), one historical reference, one contemporary practice or scene, one NON-ART reference, and what to explicitly avoid. Flag any reference you're not certain actually exists.
The research mapperI'm researching [topic] for [project]. Give me the lay of the land: key traditions, 5 names I should know, 3 tensions or debates in the field, and what a beginner typically gets wrong. Flag anything you're not certain is real. Then list the 5 questions I should be asking for this specific project.
The cross-domain jumpMy project is about [feeling/structure — e.g., "slow accumulation, then sudden collapse"]. Name 6 NON-ART domains where this same pattern appears (science, industry, nature, sport, craft), and for each: one visual phenomenon from that domain I could research as reference.
The verification interrogationList every artist, work, movement, and factual claim in your last answer. For each: how confident are you it's real and accurately described? What exact search would confirm it? Mark anything that might be invented or blended from multiple real things.
The anti-cliché passHere is my moodboard/territory description: [paste]. Which elements would appear in ANY student's board for this brief? Replace each cliché with a more specific alternative drawn from my three sources: [list].

07Lab: Triangle → Board → Statement (70 min)

Cold start friendly. No project? Use the standing brief: identity for a Dublin venue hosting both trad sessions and electronic nights.

Part A — Triangle hunt (30 min)

  1. Run the territory expander once; pick the territory that surprises you most.
  2. Build your triangle: historical + contemporary + non-art.
  3. Verify all three exist (SIFT + image search); source list with [real — verified] marks.

Part B — Mini concept board (25 min)

6–10 items showing a direction, not a pile. Real references (source noted) and your own sketches/photos welcome. AI-generated items allowed only if labelled [AI-generated] — and outnumbered by real ones.

Part C — Direction statement (15 min)

150 words, written by you: the direction, its named sources, what it refuses, why it fits. Optionally run the anti-cliché pass after. Final wording is yours.

Zero-generation track

This lab needs no image generation at all — by design. Use AI only for territory text and question-finding, or skip AI entirely and build from archives. Verification and synthesis are identical. Equal credit.

08Integrity Check

This week's four
  1. Could you narrate the path from sources to direction in a crit? If not, you don't have a direction yet.
  2. Is every "fact" on your board verified? One invented reference poisons the well.
  3. Is your contemporary source engaged with, or imitated? Would that artist recognize respect or theft?
  4. Are AI-generated items labelled? An unlabelled synthetic reference is a lie you're telling yourself.

09Troubleshooting

ProblemFix
All three territories still feel clichéYour feeling-word is still an aesthetic name. Go specific: not "cozy" — "the 20 minutes after rain when everyone reappears"
Reference checks out but is boringVerified ≠ chosen. Verification filters; taste selects. Keep hunting
Deep research report is impressive but huge"The 5 claims most relevant to my brief, with sources" — then SIFT those five only
Can't find a non-art sourceRun the cross-domain jump; pick the weirdest result and look at real photos of it
Grounded tool contradicts the chat modelGood — that's the system working. Trust the grounded one, then verify it too
"Everything's been done" despairWatch Everything is a Remix. Combination IS the originality; three points make territory

10Resources

Why this video: the definitive hour on influence vs. copying — copy, transform, combine. The final third addresses AI art directly. Watch it before you next feel guilty about references.

Optional extension

Next week: Image Generation and Editing → — bring this week's board; it becomes your image brief.