Make an AI Movie
AI Movie Prompt Log Template: Track Every Shot and Version
An AI movie prompt log is the production record that connects a planned shot to every generation attempt and the clip that finally reaches the edit. It is not another place to brainstorm prose. It captures the exact prompt, tool context, references, settings, output, evaluation, and decision so a team can compare versions without relying on memory or filenames such as final-final-3. A well-designed log saves time when a shot must be revised weeks later, a model changes, or the editor needs to understand which variant is approved. This guide provides a practical template and workflow that complements a shot list without duplicating it.
1. Give the Prompt Log One Clear Production Job
The shot list defines what the movie needs; the prompt log records how each attempted asset was produced and judged. Use the same immutable scene and shot ID in both documents, then assign a separate attempt ID to every generation. A shot such as 024-030 might have attempts A001 through A008, each tied to one output or batch. This structure keeps the editorial purpose stable while prompts and settings change. It also prevents a successful experiment from becoming detached from the scene it was meant to serve.
Choose a system that the team can update immediately: a spreadsheet for a small film, a database-style table for many collaborators, or a production tracker with linked assets. Keep one row per attempt rather than one endlessly overwritten row per shot. Store large media files in an asset repository and place durable asset IDs or links in the log. Define the owner of each entry and the point at which it becomes read-only. The log succeeds when a collaborator can open an approved clip and trace it back to its full generation context in under a minute.
2. Copy These Core Fields into the Template
Start with identity and context fields: project, sequence, scene ID, shot ID, attempt ID, creator, generation date and time, story function, priority, and requested screen duration. Then record the generation environment: provider, model, model version when exposed, generation mode such as text-to-video or image-to-video, interface or API, aspect ratio, resolution, source duration, frame rate when selectable, seed when available, and every other nondefault control that could affect the result. Never place model or setting details only in a freeform note, where they cannot be filtered later.
Add input and output fields: full positive prompt, negative prompt or exclusion instructions when used, reference asset IDs, first- and last-frame IDs, source media, character and location version, batch size, output asset ID, preview link, storage path, and technical processing performed after generation. Finish with decision fields: status, selected take, reviewer, review date, strengths, defects, continuity notes, edit range, rejection reason, next test, and approval level. Include a rights-and-provenance note for third-party references or licensed inputs, but treat the log as supporting documentation rather than a substitute for the production's clearance records or legal analysis.
3. Version Prompts with Controlled Experiments
Save the exact submitted text before generating and never silently replace it. Give the initial prompt a version such as P1.0, use P1.1 for a small revision, and start P2.0 when the shot concept or staging materially changes. Add two short fields to every revision: “change made” and “expected effect.” For example, change made might say “replaced orbit with locked camera,” while expected effect says “preserve background geometry and emphasize performance.” These notes turn iteration into a test instead of a succession of increasingly complicated prompts.
Change one major variable at a time when diagnosing a problem. If the prompt, reference image, model, duration, and motion control all change together, the resulting improvement cannot be attributed to a useful decision. Duplicate the prior attempt, alter the target variable, and link the two rows as a comparison pair. When a batch returns several outputs from identical inputs, keep one prompt version but assign each clip a distinct output or take ID and evaluate them separately. A seed or saved setting may improve repeatability in some systems, but it should be logged as context rather than treated as a guarantee of identical future output.
4. Review Outputs with a Shared Scoring Rubric
Evaluate the clip against the shot brief before reacting to surface polish. A compact rubric can score story clarity, character and wardrobe continuity, action accuracy, performance, camera behavior, environment consistency, technical integrity, and editability. Keep the scoring scale small and define each value so reviewers apply it consistently. Pair scores with one observable note: “prop changes hands before the reaction” is more useful than “timing feels wrong.” Mark hard failures separately from preferences, because a continuity break may make a clip unusable while a slightly different cloud pattern may not matter.
Use approval levels that state what the clip is approved for: exploration, temp edit, final candidate, or final. Record the approving role and date, and require a new approval when the underlying file is regenerated, upscaled, retimed, or materially altered. Comments should stay on the attempt row or link to a preserved review thread, not disappear into private messages. When opinions conflict, return to the shot's story function and the cut. The correct selection is the version that serves the sequence, not automatically the most impressive standalone generation.
5. Connect the Log to the Edit and Learn from It
When an attempt enters the timeline, record the edit project version, clip name, in and out points, retiming, crop, stabilization, compositing, color, and audio treatment. Keep the raw generation immutable and create derived asset IDs for processed versions. If the editor replaces the clip, update the selected-attempt field without deleting the earlier decision. This preserves a reliable trail from source generation to final frame and makes it possible to rebuild a sequence, prepare delivery notes, or revisit a shot without guessing which file was used.
Review the log at sequence milestones. Filter rejection reasons to find recurring problems, compare attempts per approved shot, and identify references or settings that consistently support the film's continuity. Convert repeated lessons into production rules, but do not mistake one model's behavior for a permanent law. Archive the final log with the locked cut, source assets, and tool-term records, while restricting any sensitive references or personal data to the people who need them. The template creates value when it shortens the next decision and leaves the finished movie with an intelligible production history.
Our guides distinguish current capabilities from forecasts and are updated as tools, policies, and industry practice change. Read our editorial policy.