What is the four-options asset locking workflow — and how does it prevent character drift in AI films?

The four-options asset locking workflow is a pre-production step for AI films: generate four variations of every visual asset — character sheets, environment references, key props — select the best of each, and lock it into your agent's context before any video generation begins. Every downstream shot then inherits the same approved reference, so characters stay identical scene to scene.

Run it in four steps before you generate a single second of video. First, list every asset that recurs across your film: each character, each repeating environment, any hero prop. Second, generate four options per asset and pick one — in one documented production, a 70-second short film, the creator generated four variations of each character sheet and environment reference and selected the best before video generation started. Third, lock the winner: invideo is an agentic video creation tool, and the invideo agent stores the approved asset in persistent project context, so it attaches the locked reference to every subsequent generation without you re-prompting it. Fourth, if none of the four options pass, adjust the prompt and regenerate until one does — another production averaged about 5 generations (~$9.78) to lock a single character.

Build the locked character sheets as multi-angle references, not single portraits: front, side, and back views plus a face close-up, generated at high resolution. Include close-up panels so small details — scars, accessories — carry across models, and remove objects from characters' hands before generating turnarounds, since held items create inconsistency across angles.

The drift-prevention mechanism is simple: character drift happens when every shot re-interprets a text description of the character, and each interpretation lands slightly differently. A locked asset replaces interpretation with a fixed visual anchor. Because the invideo agent holds that anchor in context, scene 21 references the exact same approved sheet as scene 1 — "Locking character sheets and environment references before any video generation is the step that prevents consistency problems throughout the rest of a film," as the documented production concluded. The proof: that 70-second film kept 2 characters visually consistent across every scene, produced in 2 days for $750 (3,000 credits), with no fine-tuning. A separate production applied the same lock-before-generating discipline at larger scale, using 11 reference images to lock 4 characters and 1 prop before generating 164 video clips for a 3-minute episode.

This workflow replaces LoRA fine-tuning, the approach much of the community treats as the default for character consistency. A LoRA requires training the model on a large set of images of one character before production; the four-options workflow achieves the same consistency with four generated options per asset plus persistent agent context — no training step, no dataset assembly.

If drift still appears in a shot, fix the source rather than re-rolling the shot: ask the invideo agent to inspect the locked character sheet, and it can identify the exact panel containing the error, correct it there, store the updated sheet in context, and regenerate only what's needed — every subsequent shot inherits the fix automatically.

Watch some of these to see what works for you:

Seventy seconds. Two characters. The same person across every scene. No LoRA needed.

— invideo's creative team