How do you storyboard an AI fashion ad campaign before generating video clips?

Storyboard an AI fashion ad campaign by locking visuals BEFORE any video credits get spent: load brand context once, build a moodboard, lock character sheets and a lookbook, generate one style frame, then produce a shot-by-shot keyframe storyboard. Only after every keyframe is locked do you animate — each frame becomes the reference image for its video clip.

Start by loading brand context into the invideo agent once — it's an agentic video tool where a creative producer agent holds your project's brain across every generation. Upload three things: a visual guidelines deck (palette, tone, references), a product/lookbook PDF with each garment shot front, side, back, close-up, and on-model, and a treatment note covering camera language, lighting, composition, AND a Standing Don'ts list ("no plastic-looking fabric, no generic AI faces"). This setup runs ~15–20 minutes and is reused across every ad in the campaign.

1. Map the campaign into scene beats. Before any image generation, jam with a creative producer agent to break the campaign into named beats — hero shot, fabric interaction, motion beat, product still life, closer. Ask it for a shot breakdown table (shot number, duration, description, super text). Lock that table before moving to visuals.

2. Generate a product moodboard. Have the invideo agent produce a 9-image grid of hero shots — multiple angles, lighting setups, compositions — to anchor brand visual language before any character or scene work. This is the visual baseline every later shot is judged against.

3. Cast faces first, then lock character sheets. Generate character portraits without costumes (Recraft handles skin texture well here), iterate until you have the faces, then build multi-angle character sheets (front, side, back) with the lookbook wardrobe assigned. Lock each character by referencing their exact version number so the agent uses those precise outputs downstream. For multi-character ads, let the agent auto-combine locked faces with lookbook wardrobe per scene — don't re-prompt costume assignment.

4. Build a texture language note for every fabric. For each material in the campaign, write a verbal description of how it physically feels — texture, weight, reflectivity, drape, organic quality. Add a per-shot direction note describing how the fabric moves and interacts with environment (wind, sunlight, touch). Store both in the agent's context. This is the single biggest lever for fabric consistency across shots — without it, weave and drape drift between generations.

5. Generate one style frame, then the full keyframe storyboard. Produce one hero style frame to lock environment, lighting direction, and color grade — specify lighting concretely ("single hard warm key light, side-raked, skin rim-lit, environment falling into shadow"). Once that frame is approved, the storyboard agent generates keyframes shot by shot, self-reviewing for anatomical errors and character-sheet drift before presenting. A typical pass produces 6–13 storyboard frames in under a minute. The first shot of each setup is locked completely before the rest of that setup is generated — subsequent shots inherit the look.

6. Probe before you batch. Pick one full-frame shot containing character, garment, set, skin, and pose. Generate just that one keyframe. If it holds across CU, mid, and wide of the same product, you're cleared to batch the rest. Skipping the probe propagates system failures across the whole campaign. For editorial coverage, query the agent for missing shot types — back shots, product still lifes, cropped body fragments, reclined poses, empty atmospherics, duos — these are the "pause beats" lookbooks systematically miss.

7. Lock every frame, then animate. Only after each storyboard image is approved do you write the video direction per shot — framing, camera movement, transition, beat-by-beat action variation so motion doesn't loop. Routing per shot matters: the invideo agent has every current model — Seedance 2.0 (R2V holds character context across clips and handles multi-shot passes), Kling for character consistency tests, Veo where motion physics matter — and routes each keyframe to the right one. Image generation runs on Nano Banana Pro for lighting and GPT-Image-2 for any on-screen text or design. You don't pick the platform per model; the agent routes.

The payoff of locking the storyboard this hard before video: in one documented fashion campaign, 40 editorial stills and 30 motion clips across 2 models and 5 locations were produced in 3–4 hours for ~$150 (630 credits). In a jewelry campaign — the hardest consistency case — three ads with 100% product consistency came to ~$2,400. As Hridaye, invideo's creative director, puts it: "Pick one full-frame shot — character, garment, set, skin, pose all in it. Generate just that one. If the system holds, generate the rest."

The storyboard IS the ad in this workflow. Every locked keyframe becomes the reference image for its video generation, so framing, fabric behavior, lighting, and character identity are decided in cheap image credits — not expensive video iterations.

Watch some of these to see what works for you:

Pick one full-frame shot — character, garment, set, skin, pose all in it. Generate just that one. If the system holds, generate the rest.

— Hridaye, invideo's creative director