AI Image and Video Generation for Marketing Teams: A Playbook

Generative tools don't replace a creative director or a brand designer. They change the expensive middle of a campaign: the rounds of concepts, the dozen ad sizes, the regional versions and the quick video cut someone asks for on Friday afternoon.

This playbook covers four jobs where AI images and video pay their way for marketing teams, the guardrails that keep output on brand and legal, and how to budget credits so nobody is surprised at the end of the month.

Before anyone books a shoot, generate visual directions for the same idea: a minimal studio look, a warm lifestyle scene, a bold graphic treatment. Show three or four side by side in the kickoff and let stakeholders react to pictures instead of adjectives.

In the Image Lab, Nano Banana 2.1 is fast and inexpensive for exploration (12 credits at 2K). Switch to GPT Image 2.5 when a concept needs a headline or packaging copy rendered in the image.

Split tests only teach you something when the variants differ in a meaningful way. Use AI to vary one element at a time: the setting (kitchen vs office), the light (morning vs night), the person (no person vs a fictional model), the product angle.

Keep the product itself identical by starting every variant from the same real product photo as a reference, and check each output against the real item. Multi-Shot turns one reference into several cinematic variations at 16 credits per shot, and Product templates give you finished scene styles at 18 credits per image.

A campaign that runs in several markets often needs different settings, seasons or signage. Keep the composition and the product, change the environment in the prompt (“the same scene in a Melbourne laneway café in winter light”), and add localized copy in your design tool so translations stay exact.

If on-image text is part of the design, GPT Image 2.5 renders quoted text accurately enough for drafts, but proofread every word before anything is published.

Most short ads start from a still. Take an approved key visual into the Video Lab and animate it: Kling 3.0 gives controlled 5- or 10-second motion (60 credits for 5 seconds at 720p), and Seedance 2.5 generates longer clips with native audio (110 credits for 5 seconds at 720p). Finish with Video Upscale to 1080p for 16 credits.

For formats with a presenter, our guide to AI UGC ads covers scripts, lip sync and the disclosure rules, and product videos from one photo covers object-led spots.

A shared prompt library. Keep a document with the prompts, models, resolutions and aspect ratios that produce on-brand results, plus reference images in shared folders in My Library. New team members start from what already works.

One model per look. Each model has its own visual character. Pick one for each campaign look and stay with it, so the series doesn't drift.

Truthful ads. In the US, Canada and Australia, the FTC, the Competition Bureau and the ACCC prohibit misleading advertising, whether or not AI made the image. AI people must not be presented as real customers or endorsers, and realistic AI content should carry the labels each ad platform requires.

Consent. Never use a real person's face, voice or likeness without permission.

Confidential launches. Read section 8 of the Terms of Service first: you own your outputs, subject to model licences, but Spicy Studio may use generated content for service improvement and marketing unless you opt out.

One account has one credit balance, so give the account owner a monthly allowance. A realistic month for a small team: 60 concept and variant images on Nano Banana 2.1 at 2K (720 credits), ten 4K upscales (100) and four 5-second Kling 3.0 clips with 1080p upscales (304) comes to 1,124 credits, inside Spicy Pro's 1,800 for €25. Ultimate (3,800 credits, €50) and Creator (7,200, €80) cover heavier video use. Compare plans on the pricing page.