Yamdu names the models behind its script-breakdown AI

The shot

Yamdu, the Munich-based production-management platform built by Seriotec GmbH, documents a beta test of AI-assisted screenplay analysis that its own AI Info page says began on 28 October 2025, backed by an unusually detailed Use of AI disclosure page describing exactly which models power which feature.

What the documents show

The Use of AI page states the software “provides AI-assisted production planning capabilities for screenplay analysis” and lists four capabilities: generating one-line scene summaries, analyzing changes between script revisions, generating production synopses, and suggesting tags for breakdown elements such as characters and props. The same page states plainly what the features do not do: they “do not generate production media assets, synthetic performances, imagery, audio, or video.” Yamdu names the underlying models — GPT-5-nano, routed through Microsoft Foundry, for the summary, revision and synopsis features, and GPT-5.4 mini for breakdown tagging — and states the AI services run in Microsoft Azure's EU regions, with customer screenplay data used only to produce the requested output and not to train the models. The AI Info page separately states the beta covering breakdowns began 28 October 2025, and the disclosure page describes each feature's controllability as “preset automation” rather than open-ended generation.

The workflow

For a production using Yamdu, the documented effect sits at the breakdown stage: a script gets an AI-suggested first pass at tagging characters, locations and props, which a coordinator can accept, edit or reject before it feeds the schedules, call sheets and budgets Yamdu also manages. The one-liner, revision-comparison and synopsis features work the same way, producing a draft a human reviews rather than a final document. Yamdu's own page states the AI capabilities “can be enabled/disabled independently on a company or project level,” meaning a production can decline the feature entirely and keep the same breakdown-to-schedule pipeline running on manual tagging.

What the tool does not change

Yamdu's own disclosure is explicit that these tools extract and summarize information already present in a script rather than inventing story or media content, and it does not claim the tagging suggestions are final; a human still confirms that a suggested prop or location tag is correct before it reaches a call sheet. This is a narrower claim than “AI production management,” and the documentation itself draws that boundary.

• Does the vendor disclose which underlying model performs a stated AI feature, and where customer data is processed?
• Is an AI-suggested breakdown tag presented as a draft a person reviews, or as a final entry the schedule depends on?
• Can the feature be disabled per project, and does that option matter for a production with its own data-handling requirements?

Yamdu's own transparency page is more specific than most vendor AI claims in this pack, naming models, data handling and what the features explicitly do not do — a useful baseline against which a vaguer claim elsewhere can be measured.