# AI Documentary Production Workflow: Sources to Final Cut > Build a factual AI documentary workflow that preserves sources, uncertainty, thesis, measured narration, reconstruction boundaries, review, and final cut. Updated: 2026-07-11 Audience: Documentary creators, producers, agencies, and AI filmmaking teams Canonical: https://onira.studio/guides/ai-documentary-production-workflow ## Key takeaways - Research and story commitments must precede expensive generation. - Measured audio should determine the visual timeline. - Each stage needs an explicit review contract and durable artifact. ## 1. Turn the topic into an editorial brief A topic such as Pompeii is not yet a film. The brief should name the audience, central question, thesis, period, geographic scope, emotional tone, target duration, and claims that require special care. A useful brief also says what the film will not imply. For factual work, attach or identify a compact set of sources before story generation. The goal is not to dump research into a model. It is to establish a small evidence boundary that the story can use and the creator can later inspect. - Audience, question, thesis, and intended change in understanding. - Source pack and uncertainty notes. - Visual, legal, sensitivity, and disclosure constraints. ## 2. Approve the story before producing assets The story blueprint should define the hook, acts, turning points, key people and places, evidence commitments, and ending. Review it before narration, images, or video consume substantial time and credits. This is the cheapest point to correct the promise of the film. The screenplay then translates those commitments into spoken narration and scene intent. Each beat should connect a line of audio to a visual purpose rather than merely requesting attractive footage. - Approve the argument and structure first. - Separate documented fact from interpretation and reconstruction. - Give every scene a narrative job. ## 3. Lock and measure narration Generate or record narration after the script has passed factual and editorial review. Check names, pronunciation, emphasis, pacing, and pauses. Measure the final audio segments so the shot plan is built against real durations. Audio-first production prevents a common failure: generating a collection of clips and then forcing the voice track to fit them. Instead, the visuals conform to the finished verbal story, which gives the edit a stable spine. - Review text before voice generation. - Check the rendered voice, not only the script. - Use measured segments as timeline constraints. ## 4. Direct stills before motion For each shot, describe what the image should look like separately from what should move. Establish character, subject, location, wardrobe, material, weather, lens, framing, and light in the still-image direction. Iterate on that lower-cost anchor before asking a video model to animate it. Motion prompts should stay economical: one main action, one camera behavior, and environmental movement that supports the line. Current generation systems are strongest with short directed shots, so long-form work still requires planning and assembly across many clips. - Appearance belongs in the image prompt. - Movement belongs in the video prompt. - Use reference sheets and accepted frames to reduce drift. ## 5. Assemble, review, and record intervention Combine narration, shots, music, sound effects, captions, and transitions on a canonical timeline. Review the complete film at normal speed and again for factual visuals, text artifacts, continuity, audio balance, caption accuracy, and disclosure-sensitive scenes. Keep an intervention log: what was generated, selected, regenerated, corrected, or edited outside the system. That record makes case studies credible, helps estimate future work, and prevents a polished excerpt from being mistaken for an unchanged first cut. - Review the whole film, not only individual assets. - Log corrections, regenerations, and external edits. - Export the final MP4 only after editorial sign-off. ## Publication checklist - The brief contains a question, thesis, audience, and source boundary. - The story blueprint is approved before asset generation. - Narration is reviewed and measured before scene fan-out. - Still appearance and motion direction are separated. - The complete timeline receives factual, quality, rights, and disclosure review. - Interventions and corrections are recorded. ## Sources - [Runway guidance for creating longer films](https://help.runwayml.com/hc/en-us/articles/26871350018835-How-to-create-longer-videos-and-films) - [Google Veo video-generation documentation](https://ai.google.dev/gemini-api/docs/video?hl=en) - [YouTube synthetic-content disclosure guidance](https://support.google.com/youtube/answer/14328491?hl=en) ## Questions ### Can one AI model make a complete documentary? A complete documentary usually requires several capabilities: research and writing, speech, image generation, short video generation, music, captions, and timeline assembly. The production system coordinating them is as important as any single model. ### Why generate narration before visuals? The real narration duration determines pacing and shot length. Measuring it first lets the visual edit conform to the story instead of stretching or compressing speech around arbitrary clips. ### Where should human review happen? At minimum, review the brief and sources, story blueprint, final narration, representative visual references, and complete final cut. Sensitive or high-risk claims may need additional specialist review. ## Product boundary - Onira delivers a final MP4; it does not upload or schedule posts on YouTube or social platforms. - Onira provides a reviewable production workflow; creators remain responsible for approving the story, facts, rights, disclosure, and final publication. - Director chat is limited to regenerating one selected PREVIEW timeline video clip; other available Studio controls are separate direct actions. - Creators must review facts, sources, rights, realistic-synthetic-media disclosure, and platform policy before publishing. - Onira does not guarantee YouTube monetization, reach, factual accuracy, or legal clearance.