Motion graphics vs AI video comes down to one question: does the video have to show something exact, or something that looks real? Motion graphics are designed type, screens, numbers and logos that move, set as real text, so they suit anything with words, prices, UI or a claim that has to be correct. AI video is footage a model generates from a prompt or an image: people, places, light, product shots, camera moves. It suits mood and physical motion, and it's the wrong tool the moment a single letter matters.
Most marketing videos need a bit of both. The practical answer is often a split: generated footage for the plate, motion graphics for the words.
Oakgen is our product, and it has a tool on each side: Oakgen Motion builds motion graphics from a brief, and the AI video generator makes footage. They are separate tools, and this guide says where each one stops.
Start with the words that must be right
Describe the video, add your brand and screens, and play a free storyboard. The exact price is on the Make video button before anything is charged.
The short answer, side by side
| Motion graphics | AI video (generated footage) | |
|---|---|---|
| What it is | Designed elements that move: type, UI screens, charts, icons, logos | Footage predicted by a model from a prompt or a starting image |
| Text | Real type: spelled right every frame | Drawn as pixels: can warp, misspell or change between frames |
| Your product | Your actual screenshots and product photos | A plausible version that may invent details |
| People and places | Not really; it has no camera | Its main strength |
| Fixing a typo | Edit the word, render again | Regenerate the clip and recheck everything else in it |
| Best for | Explainers, announcements, quote cards, lower thirds, claims | Mood films, ad hooks, b-roll, product beauty shots |
| Typical failure | Looks like a slide deck when overloaded with text | Drift, hallucinated details, unusable text |
If you want the wider map of tools that make motion graphics (editable tools, code renderers, script-to-video builders), the hub post on what AI motion graphics can and can't do covers the four families. This post stays on the narrower choice: for this particular video, footage or graphics?
Two plain definitions
Motion graphics are graphic design with timing: a headline on the beat, a counter ticking to the right number, a logo resolving at the end. Every element was placed on purpose, so the words are correct and the screens are yours. Kinetic type is the text-led subset, covered in the kinetic typography guide.
AI video here means generated footage from models like Veo, Kling or Seedance. Nothing in the clip was placed; it was predicted. That's why it can show light on a wet street, and why it can't promise the word on the cup is spelled right. To compare footage tools, see Runway alternatives, Higgsfield vs Runway or Sora alternatives.
An animated explainer sits on the motion-graphics side even with illustrated characters: it's designed, not predicted. If you'll need to change it later, that's what decides it, not the style.
Job by job: which approach fits
Each row is a common marketing video and the split we'd pick between footage and graphics. For which kind of motion-graphics tool suits each job, use the decision table in the hub post.
| Job | Use | Why | Watch for |
|---|---|---|---|
| Launch teaser or feature announcement | Motion graphics; at most one generated shot | The value is your real screens and exact names, dates and URLs; a footage shot only adds place | A generated hero shot that invents a version of your product |
| Product demo | Screen recording or real footage, with motion-graphics callouts | A demo has to show the real thing working; generated UI is fiction | Any AI shot that could be read as the actual product |
| Ad hook (first seconds) | Either; test both | Footage wins when the hook is a visual surprise, kinetic type wins when the hook is a claim | Generated text in the hook frame |
| Talking-head replacement | Motion graphics with a voiceover | Delivers the information without a face; a generated speaker is a different tool | Pretending a generated person is a real customer |
| Mood or brand film | AI video, ending on a type card | Footage carries the feeling; the only exact thing is the closing line, so set it as type | Logos or signage the model draws into the scene |
| Explainer | Motion graphics, a footage moment if needed | Steps, diagrams and numbers need structure and exact words | Text-heavy scenes held too briefly to read |
| Social quote card | Motion graphics | The words are the content; misspelled generated text ruins it | Quote attribution and line breaks in 9:16 |
| Lower third | Motion graphics in your editor or an editable tool | Names must be exact and it sits on other footage; see animated lower thirds | Exports without an alpha channel when you need an overlay |
| Music visualiser | Motion graphics for anything on the beat; generated loops for texture | Graphics can be timed to the beat; generated clips are timed in the edit, not the prompt | Seams where a short generated loop repeats |
The ad hook is the one place where we wouldn't give a default. A generated shot of something impossible stops a scroll in a way type rarely does. A blunt claim in big letters ("Your invoices, sent before lunch") stops a different viewer. Make both versions of the first two seconds and let the ad platform tell you; don't decide it in a meeting.
The talking-head replacement is usually asked for because nobody wants to film. Generating a person who reads your script is avatar-tool territory, with its own uncanny-face problems and, depending on platform, disclosure rules you should read on the platform's own pages. Kinetic type with a voiceover gets the same information across, keeps every word exact, and doesn't invite the question "is that a real person?"
For full scripts, see making a product launch video in one evening and the 20-second feature announcement template.
Where AI video fails
Text inside generated footage
A video model draws letters the way it draws leaves: as patterns that look right. Short, common words sometimes survive. Product names, prices, URLs and non-Latin scripts are where it tends to go wrong, and a word that's correct in frame 10 can be wrong in frame 40. The rule is simple: don't ask a generator for any text you'd be embarrassed to see misspelled. Prompt for the scene without text and set the words yourself afterwards.
Drift and invented details
Across a clip, and much more across several clips, things wander. A logo gains a stroke, a bottle's label changes, a phone grows an extra camera, a person's face shifts between shots. For mood work that's tolerable. For anything a customer will compare with the real product, it isn't: a generated version of your app's dashboard is a picture of an app that doesn't exist.
Check the last frame against the first, and check details like buttons, ports, labels and hands at full size.
Timing you can't specify
You can describe a camera move, but you can't tell a generator "the cup lands on the second beat of bar three". Timing happens in the edit, by trimming. That's fine for b-roll and a problem for anything that should hit a voice line or a beat exactly.
Where motion graphics fail
No footage, no people in motion
Motion graphics have no camera. They can't show your product being used in a kitchen, a crowd at an event or a founder walking into an office. If the point of the video is "look at this in the real world", graphics alone will feel abstract.
The slide-deck trap
Because adding a line of text is cheap, motion graphics tend to collect too many words. A 30-second video with eight bullet points is a slide deck with transitions. Hold times, words per second and how much to cut are covered in the kinetic typography guide; the short version is to cut the script by a third before you build anything.
Only as good as your screens
Without crisp screens, product photos and a logo file, motion graphics fall back on type and shapes: fine for an announcement, thin for a brand story.
Editability: what a late typo costs
The difference shows when someone finds a wrong price after export, legal wants a word changed, or the launch date moves.
| Change after the draft | Motion graphics | Generated footage |
|---|---|---|
| Fix a typo in a headline | Edit the word, render again | If the text is in the footage: regenerate, and the rest of the clip may change too |
| Price or date changes | Edit the number | Same as above; the reason not to bake numbers into footage |
| Swap a product screen | Replace the image | Not possible; the model never had your screen |
| New logo | Replace the file | Regenerate every shot it appears in |
| Translate into another language | Swap the words, check line lengths | Fine if the footage has no text; a rebuild if it does |
| Make a 9:16 version | Re-lay out the same scenes | Regenerate or crop, and recheck the framing |
| Change the mood of a shot | Limited; graphics have no light or place | Regenerate the shot, which is what generators are for |
The pattern is consistent: put anything likely to change in the layer that is cheapest to change. Words, numbers, logos and screens go in an editable layer. Light, place and physical motion go in footage. How each kind of tool charges is covered in the hub's cost and time picture.
Keep the words in a layer you can edit
Motion sets your words and screens as real type, with voice and captions. Change words and scene order on the storyboard before you approve a price.
The hybrid workflow: footage plate, type on top
When a video needs both, split the work by what each side is good at and join it in an editor.
1. Mark every line of the script. Next to each line, write "exact" or "real". "Exact" lines carry words, numbers, prices, UI or a logo. "Real" lines need a place, a person, light or a physical product moving. Most scripts end up with one or two "real" moments and the rest "exact".
2. Generate the plates with no text in them. In the AI video generator, prompt each "real" shot without signage, labels or captions, and ask for empty space where your words will go (sky, a plain wall, a soft background). If the shot must show your physical product, start from a real product photo rather than a description. Make a few takes of each and pick on detail, not on the thumbnail.
3. Build the type and screen scenes. Make the "exact" parts as motion graphics: in Oakgen Motion, in an editable motion tool, or in your editor's title tools. Keep each scene full-frame and self-contained.
4. Join them in your own editor. In Oakgen, this step happens outside both tools. Motion doesn't import your footage into a video made from a description: its scenes are built from type, your screen images, graphics, voice, captions and music. There's no handoff from Motion into a Compose editor or any other editor; download the MP4 and join it yourself. And its output is an MP4 without transparency, so a Motion video can't sit on top of footage as an overlay. In practice that leaves two patterns:
- Cut between them. Footage shot, then a full-frame Motion scene, then footage again. This is the cleanest join and suits ads and launch teasers.
- Type over footage. For a line of text over a generated shot (a lower third, a price on a pour shot), use your editor's own text tool. Match the font and colour to your motion-graphics scenes so the two halves look like one video.
5. Sort out sound once, for the whole cut. If you're joining pieces, lay one music track across the full edit in your editor and set Motion's sound to none, so the music doesn't restart at every join. If the Motion scenes carry a voiceover, keep the footage shots short enough that the voice doesn't feel interrupted, or write the footage moments as pauses in the script.
6. Proof the final export, not the parts. Read every word at full size in the finished file. Then watch the footage shots for drift at the cuts, where a changed detail is easiest to spot.
Invented example, for illustration only: a coffee-grinder brand making a 30-second launch ad. The first three seconds are a generated macro shot of beans falling, no text. Seconds 3 to 20 are a Motion video: the product name in kinetic type, three product photos with one line each, the launch price. Seconds 20 to 26 are a second generated shot, a pour, with "Ships in March" set over it in the editor. The last four seconds are the end card from the Motion scenes. If the price changes the day before launch, the footage is untouched. In Motion, once you've pressed Make video, the draft still lets you change on-screen words of about the same length; spoken lines are locked, so a price read out in the voiceover means a new paid render.
Five questions to pick one
Answer these in order for the video in front of you. Keep a tally.
1. Does any word, number, price, name or product screen on screen have to be exactly right? Yes: those parts are motion graphics, whatever else happens. No: go to question 2 with a point for AI video.
2. Does the viewer need to see people, places or a physical product moving in the real world? Yes: those shots are generated footage (or a camera). No: a point for motion graphics.
3. Will the words change after you publish, or across versions (price, date, language, a 9:16 cut)? Yes: keep all words in an editable layer, never inside generated footage. A point for motion graphics. No: no point either way.
4. Do you have real material to show: crisp screens, product photos, a logo file? Yes: a point for motion graphics, because your real product beats any generated stand-in. No: generated footage can fill the frame, but it must not pretend to be your product.
5. Is the job mood or information? Mood: a point for AI video, and finish on a type card. Information: a point for motion graphics, with at most one footage moment.
How to read the tally: three or more points for motion graphics and the video is type-led; build it as motion graphics and add a footage shot only if a question 2 answer demands one. Three or more for AI video and it's footage-led; generate the shots and set whatever text survives in the edit. A split tally means hybrid, and the workflow above applies. A "yes" to question 1 overrides the tally for those lines: exact things are never generated.
Where Oakgen fits, and where it doesn't
Oakgen is our product. Oakgen Motion covers the graphics side; the hub post explains how it works. Two limits matter here: videos made from a description contain no AI footage, and output is MP4 without transparency, so it can't be an overlay. The AI video generator is a separate tool for footage, with several models as of October 2026. There's no handoff between them yet; you join them in your own editor. For a real product demo, skip both and record your screen.
Frequently asked questions
Should a launch video use AI footage or motion graphics?
Mostly motion graphics, because the product name, date, URL and screens must be exact. Add one generated shot only if the launch needs a sense of place or a physical product in use.
Can I fix a typo in an AI-generated video without regenerating it?
Only if the text was added on top of the footage in an editor. Text the model drew into the frame is part of the pixels, so fixing it means a new take.
Is motion graphics or AI video better for ads?
It depends on what the ad has to prove. Generated footage can make a striking opening shot, while the claim, price and offer belong in motion graphics because they must be exact and will change between versions.
When should I use motion graphics instead of AI video?
When words, numbers, product screens or logos must be exactly right, when the copy will change after you publish, or when you need the same video in several sizes and languages.
Can I combine AI-generated footage and motion graphics in one video?
Yes, and many marketing videos should. Cut between footage shots and full-frame type scenes, or set text over the footage with your editor's title tool.
Is an AI-generated video cheaper than an animated explainer?
Per clip it can look cheaper, but count the takes needed for a clean shot and the regenerations when the script changes. For text-heavy explainers, motion graphics usually cost less over the life of the video because fixes are edits.
Can AI video replace a talking-head video?
Not directly. A generated person reading your script is an avatar tool's job, with its own quality and disclosure questions. If you only need the information delivered, kinetic type with a voiceover often does it without a face.
Make the type-led part of your video
Code-built type, your screens, voice, captions and music, 15 to 60 seconds in 16:9 and 9:16. Play the storyboard free, then approve a visible price.
