Every claim in this guide comes from a published rule that you can open and read yourself, linked in the sources at the bottom of the page. That matters here more than usual, because advice about AI festival submissions is full of secondhand summaries, and the festivals themselves are more specific than the summaries suggest. Two dedicated AI festivals have written documentation requirements directly into their 2026 rules, and one of the largest general festivals turns out to require something related without ever using the word AI.
WAIFF: a production file with every submission
The World AI Film Festival states its requirement in one sentence: “A production file explaining the use of AI tools is required for every submission.” It applies across the competition categories, so short films, feature films, micro series, and advertising entries all carry it. The festival does not publish a template for the file, which means the entrant decides what a credible account of the production looks like. The eight fields at the end of this guide are our answer to that question.
The same rules also set a floor on how much of the work must be generative: “Each submitted Short Film, Feature Film, Micro Series, or Advertising must use at least three generative A.I. tools, including one for image creation.” Read together, the two rules say something useful about how the selection committee reads a production file. They expect a pipeline of several tools, and they will look for that pipeline in your documentation. A vague file that names one tool reads as either an incomplete record or an ineligible film, and neither reading helps you.
AAIFF: disclosure at submission, with teeth
The AAIFF terms are the most demanding published requirements we found, and they concentrate on four duties. The first is disclosure of the technical pipeline: “the AI model(s) and production pipeline must be disclosed at Submission. Incomplete or false disclosure may result in disqualification.” Note the timing. The disclosure is due when you submit, not when you are selected, so the record has to exist before the deadline, not after the good news.
The second duty is credit. The terms require that all contributors must be credited and that every person who appears in or contributed to the work has consented to its submission. In an AI production that reaches further than a conventional credit roll. It includes the voice you cloned with permission, the person whose likeness shaped a character, and everyone whose consent your submission now warrants.
The third duty is a rights warranty that names AI elements explicitly. Entrants represent that they have obtained all necessary rights, consents, and clearances for all third party material incorporated in the work, and the terms list what that means: music, footage, voices, likenesses, trademarks, brand assets, datasets, and AI models. The fourth extends the warranty to the tools themselves: “The use of AI models, datasets, and tools in the creation of the Work complied with the applicable terms of service and licences of those tools and with applicable law.” That sentence is only safe to sign if you know which tools touched the film and what their terms allowed at the time.
What the big general festivals actually say
Coverage of this topic often claims that the major general festivals now demand AI disclosure. We checked the primary text. The Tribeca Festival’s published 2026 Rules and Regulations, linked in the sources below, contain no AI-specific clause at all. What they do contain is the standard warranty every large festival uses: the entrant confirms that they hold the rights to everything in the film. That warranty does not need the word AI to reach your AI elements. A cloned voice you cannot show permission for, or a generated shot built on someone else’s footage, breaches it just as surely as an unlicensed song.
The practical conclusion is the same from both directions. The dedicated AI festivals ask for the record explicitly, and the general festivals make you warrant facts that only a record can support. Either way, the film travels with an account of how it was made, and the only question is whether that account is assembled from records or reconstructed from memory.
The eight fields a submission pack needs
Every published requirement above is answerable from eight fields. This is the structure we use for a festival production file, and each field maps to a rule quoted in this guide.
- The tools and models, with exact versions. WAIFF expects at least three generative tools in the pipeline and will look for the list. Name them the way you would credit a lens.
- The pipeline, stage by stage. AAIFF asks for the production pipeline, not a tool list. One honest paragraph per stage covers references, image generation, video generation, voice, music, edit, and grade.
- The prompts and references behind the hero shots. Nobody wants all four hundred generations. Juries want the shots that carry the film: the prompt you used, the references that went in, and what changed between versions.
- The human contribution. This is the part selection committees actually score. Write it like a director’s statement: why this take and not the other twelve, what you cut, where you overruled the model.
- Every contributor, including voices and likenesses. AAIFF requires all contributors to be credited and their consent warranted, which includes the people you rarely think of as credits.
- Rights, licenses, and model terms. The AAIFF warranty covers music, footage, voices, likenesses, brand assets, datasets, and the terms of service of every AI tool used. The Tribeca warranty reaches the same elements without naming them.
- The disclosure statement. One page, honest and specific: what was generated, what was filmed, what was licensed, what was made by hand. Festivals compare it against the film.
- The final production record. Which version you submitted, who approved it, and when. This is also what a client, a broadcaster, or an EU transparency request will ask for next.
Build the record during production, not before the deadline
The teams that struggle with these requirements are not the ones with something to hide. They are the ones who produced honestly and kept the evidence in chat threads, file names, and memory, because no client had asked for it yet. Reconstructing a production file weeks after delivery is slow and unreliable, and the AAIFF terms make unreliable expensive, since an incomplete disclosure is grounds for disqualification. The cheap version of compliance is a record that grows while the work happens.
That record is what Behind The Workflow keeps. Each asset carries its tools, model versions, prompts, references, contributors, and approvals as one record, created during production and exportable when a festival, a client, or a regulator asks. For what such a record can and cannot prove about your work, read AI provenance: what you can prove and what you cannot, because the honest boundaries of the record belong in your disclosure statement too.