How do you lock your visual style before generating an AI video storyboard?

Lock visual style BEFORE the storyboard by loading brand context once, writing a treatment that names camera language, lighting, color, texture and explicit don'ts, generating one hero style frame, validating it with a single probe shot at close-mid-wide, then locking that frame plus its version number as the anchor every storyboard panel inherits.

Start by loading the project's persistent context — brand guidelines, color palette, tone of voice, product reference angles, and any existing reference films — into the invideo agent's context tab once, so every subsequent generation reads from the same brain instead of re-prompting per shot. invideo is an agentic video tool with all current image and video models routed under one project context, which is what lets a style lock actually hold across panels.

Write a treatment that defines the visual code. Spell out aspect ratio (in your film's delivery format), color palette, lighting direction (e.g. "single hard warm key, side-raked, environment falling into shadow"), film-stock/texture feel, pacing, and a depth structure if the look needs it — foreground prop plane, subject plane, midground object, soft backdrop. Include a Standing Don'ts section listing the failure modes to ban outright ("no plastic-looking fabric", "no generic AI faces"). Hridaye, invideo's creative director, calls the treatment "the most important" of the setup documents because it carries "your camera language, lighting, composition, and also a bunch of things that you don't want."

Add a texture/material language where materials matter. For fashion, jewelry, or product work, describe in words how each material physically behaves — weave, reflectivity, weight, how it moves in wind or on skin. Stored in context, this governs fabric and surface behavior across every shot without per-shot prompting.

Generate one hero style frame, not a storyboard yet. Ask the invideo agent to produce a single representative frame — the look you want every panel to inherit. Iterate on it cheaply at the image stage (lighting, palette, grade, framing) before spending any video credits. GPT-Image-2 is strong for establishing the aesthetic baseline; for product-heavy frames, run the GPT-Image-2 base through Nano Banana to lock exact product geometry into the locked aesthetic. Nano Banana Pro renders lighting better where that's the dominant factor.

Run a single probe shot to validate the lock. Before any batch generation, produce one full-frame shot that contains character, garment/product, set, skin, and pose all in it. If that holds, the system is ready to scale; if it doesn't, the style isn't locked yet — fix it before the storyboard. As a sharper consistency test, generate the same hero at three focal distances (close, mid, wide) and confirm the look reads identically across all three.

Lock the frame by version number and make it the anchor. When the style frame is right, lock it explicitly by referencing its exact version number in the lock command so the invideo agent uses that precise output as the reference for every storyboard panel that follows. One clean image works as the anchor for every future generation — characters, environments, and grade all inherit from it.

Only now generate the storyboard. Ask the invideo agent for the shot breakdown (shot number, duration, description, on-screen text) and then storyboard panels one at a time against the locked style frame and character/product sheets. Review and lock each panel before moving on — by the third or fourth panel the agent has enough taste context that notes like "more cinematic low-angle" land first or second try. If a model gets compared at the panel stage, route by job: Veo and Seedance 2.0 for cinematic motion later, Kling 3.0 where multi-shot continuity matters, all available inside the same invideo project so the lock travels with you.

The payoff of front-loading the lock is downstream: in one documented production, by the time storyboarding began the agent landed creative directions in the first or second try from shot four onward; a coverage sheet generated 20+ usable shots in three generations once the style was locked. That's the whole reason to do this work before the storyboard, not during it.

Watch some of these to see what works for you:

The third and most important one is the Treatment Note, which is things like your camera language, lighting, composition, and also a bunch of things that you don't want.

— Hridaye, invideo's creative director