AI Filmmaking Team Workflow: Shared Projects + Version Control
A quick 300-word overview on the essential techniques for ai filmmaking team workflow: shared proj...
Solo AI filmmaking is one thing. Running a team — multiple people generating, selecting, editing, and revising the same production — introduces coordination challenges that single-person workflows don't face. Without explicit systems for shared projects, naming, asset management, and version control, team productions degrade into confusion: duplicate generation, lost assets, overwritten work, and prompt archaeology ("what exactly were we doing for this shot three days ago?").
This guide covers the systems that make team AI filmmaking work: project structure, naming conventions, prompt versioning, asset locking, and review-approval workflows.
The Core Coordination Problem
On a traditional film set, the camera department knows which shots belong to them, the art department knows what they're building, and the AD coordinates the schedule. In AI production, the boundaries are less clear. A generator, a prompt engineer, a reviewer, and an editor may all be working on the same shots simultaneously without clear ownership boundaries.
The solution is explicit ownership: every shot is assigned to one person at one stage. The workflow system defines who owns what and when ownership transfers.
Project Structure
Every production should have a defined folder structure that all team members use consistently:
/[Production Name] /pre-production /character-references # All 14-image sets and turnarounds /storyboards # All shot panels with annotations /shot-list.csv # Master shot list with assignments /generation /[shot-number] # One folder per shot /attempts # All generated clips (named with attempt number) /selected # The chosen clip(s) from this shot /review /approved # Reviewer-approved clips, ready for edit /revision-needed # Flagged clips with revision notes /edit /assembly # Work-in-progress edit files /export # Final exports /audio /score # Generated music stems and masters /sound-effects # SFX files /post /color-work # Color-graded masters per shot
This structure is consistent regardless of who generates which shots. Anyone on the team can find any asset without asking.
Naming Conventions
Consistent naming is the single highest-leverage coordination tool. Every file in the production uses the same naming system:
Video clips: [ProductionCode]_[SceneNumber]_[ShotNumber]_[AttemptNumber]_[Model].[extension]
Example: ANMRPH_S03_SH012_A03_SD25.mp4 (Anamorpha production, Scene 3, Shot 12, Attempt 3, Seedance 2.5 model)
Images: [ProductionCode]_[Type]_[Subject]_[AttemptNumber].[extension]
Example: ANMRPH_CHAR_MARIA_14REFS_A01.jpg
Audio: [ProductionCode]_[Type]_[Name]_[Version].[extension]
Example: ANMRPH_SCORE_MAIN-THEME_V03.wav
Post the naming convention document in the project folder's root as NAMING-GUIDE.md. Every team member reads it before generating anything.
Prompt Versioning
Prompts are production assets and must be versioned. When a prompt is revised, both the old and new versions should be recorded with the reason for the change.
Prompt log format (maintain as a CSV or in Anamorpha's shot list tool):
| Shot | Version | Prompt | Date | Author | Reason | |---|---|---|---|---|---| | S03_SH012 | v1 | [original prompt] | 2026-06-01 | JT | Initial | | S03_SH012 | v2 | [revised prompt] | 2026-06-02 | MR | Added motion stability anchor after v1 artifacts | | S03_SH012 | v3 | [final prompt] | 2026-06-03 | JT | Switched to Kling for better cloth physics |
This log answers the question "what was the final prompt we used and why?" months later when you're generating supplementary shots or revising for a new format.
Asset Locking
When an asset is selected and approved, it must be locked. Locked assets cannot be revised without explicit team consensus.
Lock status system:
- Open: Under development; can be revised
- Pending Review: Complete, waiting for reviewer sign-off
- Approved: Locked; do not modify without team discussion
- Rejected: Needs revision; assigned back to generator with notes
Track lock status in the shot list CSV. Every team member can see the current status of every shot at a glance.
Review-Approval Workflow
The review workflow defines how outputs move from generator to editor:
Stage 1 — Generator Review (self-review): Generator scores their own output batch using the 0–3 scoring rubric (intent accuracy, technical quality, creative quality). Selects the best output and moves it to the review queue with self-score.
Stage 2 — Reviewer Review: A dedicated reviewer (not the generator) looks at the selected clip and either:
- Approves: moves to approved queue
- Requests revision: adds a note describing specifically what needs to change and assigns it back to the generator
Stage 3 — Editor Integration: Once approved, the clip is moved to the edit folder and marked in the shot list as ready. The editor pulls from approved clips only — never from the attempts folder.
Communication Norms
For team AI productions, establish explicit communication norms:
Daily standup (async): Each team member posts a brief update: "Generated shots S03_SH010 through S03_SH015, three passed review, two need revision. Waiting on character reference for Scene 5."
Revision notes are specific: Not "the face is wrong" — "the character's eyes are significantly different from the reference; increase image reference weight and regenerate."
No silent overwrites: If you modify a file someone else generated, communicate the change. Silent modification is the fastest way to break team trust.
Version your exports: When a new edit version is exported, increment the version number and add a note to the project log describing what changed from the previous version.
Scaling the Team
Different production roles benefit from explicit ownership in AI production:
- Prompt Engineer: Writes and refines prompts; owns the prompt log
- Generator: Runs batches, applies the prompt, runs self-review; owns the generation folder
- Art Director / Reviewer: Runs Stage 2 review; owns the review decision
- Editor: Assembles from approved clips; owns the edit folder
- Producer: Maintains the shot list status and coordinates between roles
Small teams often combine roles. The important thing is that someone owns each function — not that there's one person per function.
Takeaways
- Explicit ownership prevents the most common team coordination failures — duplicate generation, lost assets, overwritten work
- Consistent folder structure and naming conventions allow any team member to find any asset without asking
- Version control for prompts is as important as version control for code — record what changed and why
- The lock system (Open → Pending → Approved) controls which assets can be modified
- Review-approval workflow ensures editors only work from approved assets
- Communication norms (daily async updates, specific revision notes, no silent overwrites) keep the team aligned
Continue exploring
- Voice-First AI Directing: Director Studio, from Scene to Screen
- Character Subtext in AI Video: Render the Tell, Not the Emotion Label
- An AI Film Series Workflow: Containers, Shared Bible, and Inherited Style
- From Script to Screen in One Day: A Full AI Filmmaking Pipeline