Brand systems were built for still images. Motion is where they are now being tested. Photo via Unsplash.
Most brand guidelines were written for a world of static assets. They specify logo clear space, colour values, type hierarchy and photographic style in careful detail. Very few say much about motion: how fast things should move, how a camera should behave, what a transition should feel like. For years that was acceptable, because motion work was expensive and handled by a small number of specialists who absorbed the brand by working with it.
AI video has changed that. Marketing, social and product teams can now produce short motion pieces in an afternoon, and many of them are doing exactly that. The result is a new kind of brand risk. A logo may drift a few degrees in the middle of a clip, a colour may shift under generated lighting, or a product may quietly gain an extra button. None of these are obvious in a thumbnail. All of them erode the consistency that visual communicators spend years building.
The good news is that the newest video models are built to be steered, not just prompted. This guide explains how reference-driven generation works, how to assemble references that keep motion on-brand, and how to review output before it reaches an audience.
From prompting to directing
The first generation of AI video tools relied almost entirely on text prompts. You described what you wanted and accepted whatever came back. For brand work, that was never going to be enough: words are a poor way to describe a specific shade of blue, a particular product silhouette or the feel of a house typeface.
Current models accept references alongside the prompt. Images describe identity and look. Short video clips describe motion and pacing. Audio describes rhythm. Keyframes pin down exactly where a shot begins and ends. Some models accept dozens of references in a single request. ByteDance’s Seedance 2.5, for example, can take up to fifty: thirty images, ten video clips and ten audio files, along with first-frame and last-frame images. That shifts the designer’s job from writing clever prompts to curating the right inputs, which is a far more natural task for anyone trained in visual communication.
Give every reference one job
The most common mistake is to treat references as a mood board and throw in everything that feels right. Models respond better when each input has a clear purpose. Think of references in six roles.
A simple way to organise references so each one does a single job.
- Identity images show the things that must not change: the product, the logo, the mascot, the wardrobe. Use clean, well-lit images on simple backgrounds.
- Style frames describe the look: palette, contrast, grain, lighting direction. Two or three strong frames work better than ten loose ones.
- Video clips describe motion: how fast the camera moves, how elements enter and exit, the pacing of cuts.
- Audio sets rhythm. A music bed or voice-over helps the model time movement to beats.
- First and last frames fix the opening composition and the landing, such as an end card with the product and logo in their approved positions.
Keep a named reference library for each campaign and reuse it. Consistency across a series comes less from prompts than from feeding the same approved inputs into every shot.
Translate your brand guidelines into motion rules
Before generating anything, write a one-page motion addendum to your brand guidelines. It does not need to be elaborate. A few decisions go a long way:
- Speed. Is the brand calm or energetic? Define a default camera speed and transition length.
- Camera behaviour. Locked-off and precise, or handheld and documentary? Choose one default and one exception.
- Light. Specify the key light direction and colour temperature used in your photography, and ask for the same in video.
- Type. Decide that titles, prices and legal text are always added in the edit, never generated inside the frame.
- Endings. Define one or two approved end-card compositions and use them as last-frame references.
These rules become the backbone of your prompts and your review checklist, and they make it much easier for colleagues outside the design team to produce work that looks like it belongs.
Review like a designer, not a viewer
AI video is easy to approve at a glance because it moves and looks polished. The problems hide in single frames. Review every piece with the same rigour you would apply to a print proof.
A short review checklist catches most brand problems before publication.
Scrub slowly through the clip and pause on the moments where the product or logo is most visible. Compare the colour of the first, middle and final seconds against your swatches, since generated lighting can drift over a long shot. Check that any text you did add is legible at mobile sizes. And watch the clip once with the sound off and once at full screen, because problems that disappear in a feed are obvious on a large display.
Budgeting for iteration
Iteration is the hidden cost of AI motion work. A single approved clip may take five or ten generations to reach, especially when a brand has strict rules. Pricing for most current models is per second of output and rises with resolution, so the smart approach is to explore at low resolution and only render finals once a direction is approved. As a reference point, Seedance 2.5 API pricing on APIMart starts at under ten cents per second for 480p drafts, with 720p and 1080p rates for final output. That makes it realistic to explore several directions for a campaign without a large budget, as long as the team is disciplined about when to step up resolution.
| Stage | Resolution | Goal | Who signs off |
| Exploration | Lowest available | Composition, motion, pacing | Designer |
| Direction | Mid (720p) | Colour, product accuracy, transitions | Creative lead |
| Final | Highest needed for the channel | Polish, end card, type added in edit | Brand owner |
Disclosure, rights and trust
Visual communicators are guardians of trust as well as style. Several platforms now require AI-generated video to be labelled, and audiences increasingly expect it. Build disclosure into the publishing checklist rather than leaving it to memory. Use only references you have the rights to, including photography, footage and music. Avoid generating realistic people who could be mistaken for real customers or employees unless you have their consent. And never let generated footage stand in for evidence, such as showing a product doing something it cannot actually do.
Good motion design has always been about restraint. AI makes it easy to produce more; the designer’s job is to make sure everything that ships still looks like it came from the same hand.
Where to start
Pick one recurring format, such as a five-second social loop or a product reveal, and build the motion rules, reference library and review checklist around it. Produce ten pieces, review them against your existing brand work and adjust the rules. Once that format is reliable, extend the system to the next one. Teams that work this way find that AI video becomes another well-governed channel in the brand system, rather than a source of quiet inconsistency.
Motion is no longer a specialist luxury. With the right references and a clear set of rules, it can be as consistent and recognisable as everything else a brand puts into the world.