GEN-04 · Artist attribution under AI assist

Ship a collaborative piece that credits human and machine roles without erasing either — and write the attribution line a wall label or portfolio header would actually carry. The piece is half the deliverable. The attribution that survives contact with a stranger is the other half.

Teacher approves collaborators (human), tools, and exhibition surface before Checkpoint 1.

• A real collaborative constraint: at least two human roles (e.g. concept + finish, or image + sound), plus a named generative step that materially affected the result.
• A tooling statement kept live from first experiment, not reconstructed.
• An attribution line drafted in three lengths: wall label (~20 words), portfolio header (~40 words), and colophon (paragraph).
• Consent for any likeness, voice, or classmate contribution, following GEN-02’s bar even if the student has not taken GEN-02.
• Public references as needed: Copyright Office AI guidance (https://www.copyright.gov/ai/); C2PA if the toolchain writes credentials (https://c2pa.org/).

Crediting only the humans, or only the model. Listing every plugin as equal when three decisions actually shaped the work. Using a collaborator’s sketch as a prompt seed without naming them. Publishing before the short wall-label form is readable by someone outside the class. Claiming the attribution “clears” copyright questions.

Five pieces: (1) the finished collaborative work on the named surface; (2) live tooling statement; (3) three attribution lengths, with the wall-label form tested on two people outside the class; (4) a role map — who decided what, including where the model’s output was accepted, edited, or thrown out; (5) refusal log — likenesses not used, claims not made, references not scraped.

CT: separating contribution from credit theater. CR: designing attribution that fits a real surface. CO: outsider readers are the instrument. CM: the wall label. CZ: the human collaborator who would otherwise disappear.

Working artist, designer, curator, or arts teacher reviews attribution drafts at Checkpoint 2 before exhibition. Standing instruction: reject credit that erases a human collaborator or pretends attribution equals permission. School-supervised.

R1: one piece, one surface, one attribution claim. R2: role map lets a peer see decisions. R3: what the work would be without the generative step is stated. R4: consent and unknown-training-data limits recorded. R5: wall label survives outsider read. R6: refusal log present; no “credit clears risk” claim.

(a) A stunning piece labeled “Created with AI” and nothing else. (b) A perfect colophon for a solo piece that was never collaborative.

• Week 1–2: Collaborators, tools, surface approved; consent handled; tooling log started before first generation.
• Week 4–5: Piece in progress; role map drafted; three attribution lengths in rough form.
• Week 7–8: Outsider label test; final statement; refusal log; mentor review.

Lane A immediately (portfolio, exhibition, AP/IB arts). Lane B only with a real host studio and division Internship wrapper — the brief alone is not HQWBL. No verified credit claim.