AI Asset Management: Versioning, Naming and Continuity

A short film generates dozens of assets. A series generates thousands: images, clips, voice, references, alternates and versions, all needing to stay consistent. At that scale, asset management isn't bookkeeping, it's what keeps the production from falling apart. This guide gives you a versioning model, a naming convention you can copy, and a way to keep characters and locations on-model from the first reference to the final shot.

Why AI productions need asset management

AI makes generating assets cheap, which means you generate a lot of them. Without structure you quickly face:

• Continuity drift: a character or location that subtly changes between shots.
• Lost work: the good take buried among dozens of others.
• Broken context: nobody remembers which reference, prompt or model produced which shot.
• Approval confusion: two people working from different "final" versions.

The fix is to give every asset a place, a version history, and a relationship to the shots that use it.

Organise assets as a hierarchy

A structure that holds up in production has four levels:

1. Scope: what part of the production the asset belongs to. One project-wide pack for things that appear everywhere, plus one scope per episode.
2. Category: the kind of thing it is: characters, locations, backgrounds, props, animations, audio.
3. Entity: one specific thing, such as "Mara", "The Old Mill" or "the brass key". Each entity has its approved reference images and description.
4. Asset: an individual file (an image, clip or audio take) attached to an entity, with its own versions.

When the same character entity is reused everywhere it appears, continuity stops being something you fight for and becomes something your structure guarantees. (See consistent characters for the reference workflow.)

AI asset versioning: how it works

AI asset versioning is the practice of treating every meaningful take of a generated asset as a numbered version of one entity, rather than another loose file in a folder. Your character isn't hero_final_v3_NEW(2).png, it's Hero, version 7, with versions 1 to 6 still attached, comparable and restorable underneath it.

A versioning model that survives a real production has four properties:

1. One identity per asset. A character, location, prop or shot exists exactly once. Every regeneration or style pass becomes a new version of that entity, so its whole history lives in one place.
2. Cheap, non-destructive iteration. Because AI makes retakes nearly free, you will produce far more takes than a traditional pipeline. Versioning makes that a strength: nothing is overwritten or lost.
3. An unambiguous current version. Exactly one version is current at any time, and everything downstream reads it unless a shot deliberately pins an older one.
4. Going back is a normal move. When version 9 drifts off-model, you don't regenerate from scratch: you return to version 6 and branch from there.

Two refinements matter at scale:

• Version the references too. When a character's reference sheet changes, downstream shots should say which reference version they were generated against.
• Upscales and edits are not always versions. A 4K upscale of a background is often a separate deliverable that lives next to the original, not on top of it.

A naming convention you can copy

Even with a versioned library, files leave the system: exports, hand-offs, edit bins. A fixed pattern keeps them readable:

Result: PRJ_EP01_SEQ010_SH020_char-mara_v003.png. It sorts correctly in any file browser, tells an editor exactly where the file belongs, and never lies about which take is current.

Review status belongs to the shot

It's tempting to mark an image as approved. In practice the same image can be right for one shot and wrong for another, so status works better where the asset is used. A simple set covers most productions: To do, In progress, In review, Changes requested, Approved. The asset keeps its version history; the shot records which version was picked and whether it is approved.

Why it matters most for series

A one-off short can survive loose organisation. An animated series cannot: continuity across episodes is the whole point. Structured, versioned, shot-attached assets are what let a small team keep a long-running production coherent, and onboard a new artist in an afternoon instead of a week.

How to do it in Cinemagiq

1. The hierarchy is built in. The Asset Hub is organised as Scope, Category, Entity, Asset, with versions. Scopes are the project-wide Main Pack and one per episode; built-in categories include Characters, Locations, Backgrounds and Props.
2. Entities carry their context. Each character, location or prop has its own page with a gallery, collections, a grid and a profile picture, and one asset can be linked to several entities.
3. Every generation is findable. New outputs are named after the model and time that made them, and All Generations shows everything the team has generated.
4. Versions stay together. Each asset shows its current version, a version compare dialog lets you check takes against each other, and the image Compare view puts two images side by side with synced zoom.
5. Shots pick the version. On the Production Board, each cell holds a stack of assets: a star marks the one shown as primary and a heart marks the best option. A cell can even use an older version, shown with its version badge. Review status (To do through Approved, plus statuses your admins add) lives on the cell, and comments can be pinned to a specific version.
6. Nothing slips through. Needs Attention lists assets that are in neither the Main Pack nor any episode, and Used In shows every board cell that uses an asset.
7. Hand-offs stay organised. Export a sequence's assets as a ZIP grouped by shot, by column, or in one folder.

Generation is the easy part now. Keeping it organised is the craft that scales. For the full pipeline, see our complete AI filmmaking guide.

Versioned assets, characters and continuity in one place with Cinemagiq.