How do you keep AI characters and visual style consistent across every shot of your film?

Consistency comes from locking references before generation and holding them in persistent agent context: multi-angle character sheets approved up front, a visual style block or treatment document saved to context once, explicit prohibitions against drift, and the same locked references attached to every prompt. One documented 70-second film kept 2 characters consistent across every scene this way — no LoRA required.

invideo is an agentic video creation tool with all the current video and image models available, and its agent holds project context persistently — that persistence is the mechanism behind every step below.

Lock character sheets before generating a single video clip. Build a multi-angle reference sheet per character — front, side, back, plus face and mid close-ups — and include close-up panels for small details like scars and accessories, because the model will hallucinate anything it can't see on the sheet. Remove objects from characters' hands before generating turnarounds so props don't vary across angles. Generate several options per asset and approve one before production: a documented 70-second production generated 4 options per character sheet and environment reference, locked the best of each, and kept 2 characters consistent across every scene with no LoRA — $750 total over 2 days. Budget for iteration at this stage: another production needed about 5 generations to lock each character (~$9.78 per character), and just 11 reference images covered 4 characters and a prop for a 3-minute episode. If a character's appearance evolves through the film, make a distinct sheet per story beat — one production created a new sheet for each sequence because the character picked up a new trinket in every location.

Save the visual style to the invideo agent's context once — then stop re-describing it. Two routes, depending on your film. For a referenced aesthetic, upload a large batch of style frames in a single message: one animated production uploaded 64 frames of its target style with the instruction "I want you to deeply understand this art style and save it into context for further generations," and the style held for the entire project. For a directorial style, load a structured visual-language document covering camera, lighting, palette, composition, atmosphere, and negative prompts — one production encoded 14 sections of a director's grammar; another encoded numerical precision like an 85:15 dark-to-light lighting ratio and palette hex values, so corrections reference the source ("warm yellow from the lamps only, like all the refs") instead of generic descriptors. Re-prompting style scene-by-scene is the anti-pattern: the invideo agent reads the document once and checks generated frames against it before returning results.

Write explicit prohibitions into the style block. State what the output must never be, not just what it should be. The documented animated production's style block read: "This MUST look and feel like Arcane animation — not live action, not photorealistic. Every surface has hand-painted brushstroke texture." Without the prohibition, models drift back toward their default photorealistic look over a long run of generations.

Attach the locked references to every prompt, in a fixed assembly order. One production assembled every prompt in the same 9-element order — camera spec, lens and aspect ratio, lighting source, palette, composition, atmosphere, mood register, film attribution, negative prompt — across every frame, and prefixed every generation prompt with the locked style block ("Every prompt after this started with it"). Let the prompt itself describe only the shot's action and camera; the locked sheets and style block carry the visual identity. Run generation in the invideo agent's Always Ask mode so you approve each prompt and its attached references before credits are spent. Once you have approved generations, replace your original outside references with your own extracted frames — approved panels anchor continuity more reliably than the references you started from.

Carry continuity shot-to-shot with reference-to-video. For sequences that must flow continuously, clip the end of an approved segment and re-upload it alongside your character and location references: Seedance 2.0 reference-to-video accepts the prior clip plus character and location references simultaneously, so camera movement, framing, and atmosphere carry across segment boundaries — something the extend feature and start/end-frame methods can't do, because they only see a single frame. On model choice: Kling 3.0 generates multi-shot sequences natively, while Seedance 2.0 reference-to-video carries character context across clips; all of these models run inside invideo, and the invideo agent routes each shot to the right one with your locked references attached.

Fix continuity errors at the source sheet, never by re-rolling the shot. When a detail drifts in one shot, ask the invideo agent to inspect the character sheet rather than regenerating blind: in one documented case it identified the exact panel containing the error, corrected it, stored the updated sheet in context, and regenerated only what was needed — every subsequent shot inherited the fix automatically.

Watch some of these to see what works for you:

Agent One reads your treatment doc once and keeps it loaded across every frame. The thread stays held, scene to scene. No re-explaining. No starting over.

— invideo's creative team