What is a knowledge bank locking technique and how does it keep AI agent outputs consistent across campaigns?

Knowledge bank locking is the practice of loading all brand assets — guidelines, product images, color palette, tone of voice — into an AI agent's persistent context, then locking that context so the agent applies the same brand rules to every subsequent generation without re-prompting. The lock is read at generation time, which is what prevents drift across campaigns.

Set it up before you generate anything: upload your brand guide, product images from every angle (including a close-up and a scale reference like the product held in a hand), your color palette, tone of voice, website link, and an explicit list of things the agent must never produce. In invideo, this lives in the invideo agent's context tab, which functions as the persistent memory of the whole project — one documented team seeded it with three PDFs and reused them across every ad in the project, and initial brand context setup takes roughly 15–20 minutes.

Then lock it. Once the context and product visuals are locked, the invideo agent treats them as fixed rules rather than suggestions: every image and video generation reads the locked bank before rendering, so brand colors, product geometry, and visual language hold without you restating them. You can lock at finer grain too — reference the exact version numbers of approved character or product iterations in a lock command so all downstream generations use those precise outputs, and ask the invideo agent mid-project for a status report of what's locked versus still open before spending more credits.

The cross-campaign consistency comes from the fact that the lock persists beyond a single ad. In one documented project, the second ad required zero context setup and shipped 33% faster (2 hours vs. 3) because the invideo agent already held the brand, visual language, and workflow in memory. Another production reported never re-prompting fabric behavior across an entire shoot because the locked brand context and shot direction were already stored. The consistency results scale with the technique: one jewelry campaign held 100% product consistency across three complete ads for $2,400 total, a two-ad clothing campaign held 100% fabric consistency for ~$600, and a localization run held 100% product and text consistency across 3 ads in 2 different markets.

The distinction that makes it a locking technique rather than just storage: the rules are enforced at generation time, on every output, not filed away in a document the agent may or may not consult. Community practice around AI agents converges on the same pattern — small, structured knowledge files pinned to the agent's context so identity and rules survive across sessions.

Watch some of these to see what works for you:

The context tab is basically the brain of the project. It holds all of this through every ad you build, without you repeating any of it.

— invideo's creative team