Generated Frames: Authorship and Disclosure in an AI-Assisted Projection
A candidate submits a 2.5 m wide animated projection for the school-based assessment. The three-minute loop combines 60 photographs taken by the candidate with frames produced by a text-to-image model; about 40% of the final frames are AI-generated. The candidate edited the frames in Photoshop, sequenced them and added a music loop from a free library. The 300-word rationale states: 'All imagery is original and handmade.' The school's honesty policy requires disclosure of all non-original material and process evidence. The candidate has six process screenshots and prompt logs but no record of the model's training data. One AI-modified frame contains a recognisable face of a friend who consented to being photographed for a portrait but did not know it would be transformed by AI. A classmate argues that AI is 'just a tool, like a brush'. The moderator asks whether the work can be accepted and how it should be described.
1. Assess whether the work can be accepted and what conditions must be met.
Read the worked answer
The work should not be accepted as submitted, but it need not be rejected outright. The rationale's claim that all imagery is original and handmade is false: about 40% of the final frames are generated by a text-to-image model, and the music is from a free library. The honesty policy requires disclosure, so the first condition is a corrected rationale that states the model used, the prompts, the proportion of generated frames, the editing and the music source. The second condition is consent: the friend agreed to be photographed for a portrait, not to have that image transformed by AI and projected publicly. The candidate must renegotiate consent for the AI transformation or remove or obscure the face. Third, the process evidence should be organised: prompt logs, screenshots, editing history and a clear account of the candidate's decisions. If those conditions are met, the work can be assessed as a mixed-media projection; if they are not, it is a honesty breach.
2. Distinguish tool use from authorship in this case, and respond to the classmate's brush analogy.
Read the worked answer
The brush analogy is too simple. A brush is a manual tool; its marks are directly controlled by the hand, and it carries no training data. A generative model produces outputs from statistical patterns learned from a dataset the candidate did not choose or document. That does not make the candidate a non-author, but it distributes authorship across the model, the dataset and the candidate's selection. In this case, the candidate's authorial work is visible in the 60 photographs, the choice of prompts, the editing of 40% generated frames, the sequencing and the soundtrack. The weak point is the claim of handmade originality, not the use of AI itself. The candidate should describe the work as 'a projection combining my photographs with AI-generated frames that I selected and edited', and should not claim that the model is merely a brush. The distinction matters because assessment rewards demonstrated decision-making, which can be documented.
3. Advise how disclosure would affect the work's meaning and how the rationale should be rewritten.
Read the worked answer
Disclosure changes the work's meaning from a claim about personal memory to a claim about mediated memory. If the rationale says, 'I used a text-to-image model trained on unspecified images to generate frames that I then selected and sequenced', the work can be read as an examination of how memory is co-produced by algorithms and personal photographs. That reading is more critical than the false handmade claim and could strengthen the concept of memory and loss. If the AI use is hidden, the work's meaning depends on deception, and the viewer's trust becomes the subject accidentally. The candidate should rewrite the 300-word rationale around three specifics: what was photographed, what was generated and what was decided by the candidate. The disclosure should also acknowledge the friend's consent. A short technical appendix with prompt logs would satisfy the moderator without turning the rationale into a confession.