Twenty-seven kept shots came from about a thousand generated

The shot

Protopica is a short creative work made in Guadeloupean Creole, an underrepresented language, by co-directors Manuel Sainsily and Will Selviz using early access to OpenAI's Sora video model. The piece premiered in July 2024 during Sora's alpha phase and was later shown at the Vancouver International Film Festival as what an OpenAI Forum event page describes as 'a dynamic, multisensory installation.' The directors' own account, published on the OpenAI Forum in March 2025, states the piece was built by generating a very large number of shots and keeping only a small fraction.

What the documents show

The directors' own forum post states a keeper ratio of roughly '1000/27 generated to keeper shots' — about a thousand generations for twenty-seven that made the final piece. This is a first-person, self-reported production account, not an independently audited log of every generation. A separate OpenAI Forum event page corroborates the creators' identities and describes the project as 'an artistic journey through identity and language, enriched by AI-driven storytelling,' framing that is itself promotional of the platform rather than an independent review of the finished work.

The workflow

Per the directors' own account, the process began with cultural research and validation involving Afro-Caribbean and Latin American community members, intended to check the film's language and imagery against lived cultural knowledge before and during generation. The directors then used Sora's alpha access to generate a large volume of candidate shots, described as a deliberately loose, high-volume pass rather than a single targeted request per shot. From that volume, the team curated a small keeper set, selecting shots that fit the story and rejecting the rest, then assembled the surviving twenty-seven into the finished cut. The directors describe this generate-then-curate structure explicitly as 'a collaboration' with the tool rather than a scripted, fully planned production.

What the tool does not change

The keeper ratio itself is evidence that most generations were rejected by human judgment, not published. Cultural review by community members is a step the tool cannot perform on its own; the directors describe it as a deliberate check they built into the process. The directors' own language — that they 'don't have full control, and that's exactly what sparks new ideas' — describes an editorial and curatorial role for the filmmaker across a large volume of output, not a hands-off generation.

• What does a roughly 4 percent keeper ratio suggest about the editorial labor required to reach a finished cut this way?
• How was cultural review documented, and by whom, before a generated shot was accepted?
• Would this generate-and-curate approach scale to a longer, narrative-driven production, or is it specific to a short, experimental piece?

Read as the creators' own account rather than an audited record, Protopica documents a specific production pattern — high-volume generation paired with aggressive human curation and community-based cultural review — that is distinct from claims of a single, directed shot request producing a usable result.