Video Ads Without a Video Team — Ship One This Week

You have watched your competitor's 15-second ad four times now. It is not better than your product. It is just visible, and you are not, because every time you sit down to make one you hit the same wall: you do not own After Effects, you have never used Premiere, and the one time you tried Canva's video templates you produced something that looked like a school assignment.

So you post another screenshot. Making video ads without a video team feels impossible, so most developer-founders never ship a single one — and video is where the attention is.

Who has this problem

You built the product yourself. You can reason about database indexes and race conditions, and you are not intimidated by hard things. But motion design is not a hard thing you have not learned yet — it is a hard thing with a toolchain, a vocabulary, and a set of taste judgements you have no way to check.

If you have ever opened a video editor, moved one text layer, decided it looked wrong, and closed it again without saving, this post is for you.

There is no shame in it. You are good at building systems. Video editors are not systems — they are canvases, and a canvas gives you no feedback until it is too late.

Why this problem exists

The tools available to you are built for the wrong person.

Video editors assume you have taste and a timeline. Premiere and After Effects hand you a blank sequence and infinite freedom. Freedom is exactly what you do not need. You need constraints — how long a card should hold, how big the type has to be, where the platform will draw its own buttons over your frame.

Template tools assume you want to look like everyone else. Drag-and-drop template libraries solve the blank canvas by giving you the same canvas ten thousand other people used. Change the words, change the colours, and you have a video that reads as generated, because it is.

Neither one tells you when you are wrong. This is the real gap. A compiler tells you your code is broken. Nothing tells you that your text is at 1.4:1 contrast and will be unreadable after the platform re-encodes it, or that your call to action is sitting underneath TikTok's action rail where nobody will ever see it. These failures are silent. You just ship something that looks amateur and never learn why.

That last point is worth sitting with, because it explains why "just try harder" does not work here. You cannot iterate toward quality on a signal you cannot see.

How to make a video ad for your SaaS, step by step

The Mark Platform's approach is to treat a video creative as a generated artefact with enforced constraints, not a canvas. Here is the sequence.

1. Do the brief before you touch the generator

This is the step everyone skips, and skipping it is why generated creatives come back generic. Before anything renders, you answer six things: the industry in the words a customer would use, the market, the audience specific enough to exclude someone, the objective, the platform, and — the important one — what actually works on that surface for that industry.

Two findings minimum. Not "video does well on LinkedIn." Something you could act on, like "explainers on this surface open on the problem, never the logo."

The platform refuses to generate without this. That is deliberate: a prompt fallback is useful when the missing piece is a nicety, and useless when the missing piece is the entire input.

2. Pick an objective, then a length — in that order

Most people pick a template because it looks nice. That is backwards. The objective decides the shape of the argument — awareness, consideration, conversion, proof, launch or retention — and the length belongs to the surface you are posting on, not to the template.

A 6-second bumper and a 60-second explainer are not the same argument at different speeds. They are different arguments. Ask for a length a template cannot carry and you get told which lengths it supports, rather than being quietly handed a 20-second cut you did not ask for.

3. Generate the project, not the video

This is the part that surprises people. generate_video_creative does not return an MP4. It returns a Remotion project — real React components, typed props, a render command — and you run the render on your own machine.

That sounds like more work. It is actually the thing that makes the rest possible:

• Your brand data never leaves your machine as video input. No file is uploaded to a video vendor, because no file is made on our side.
• A/B variants are prop sets, not regenerated code. Same composition, different message. That is what makes a result attributable: when variant B wins, you know the hook won, not the layout.
• You can read the source. It is React. You already know how to change a number in a file.

4. Let the constraints do the design work

The generated project has already resolved the judgements you had no way to check. Contrast is measured against the actual background and corrected if your brand colour would have landed unreadable — and you are told it happened, so you never discover we altered your palette by noticing the video does not match your brand guide. Type never drops below a floor, because people watch feed video at a third of full size. Platform safe areas are respected, so the closing card is not hidden under a Follow button. Cards are timed by how long their own text takes to read, not by dividing the duration evenly.

None of these are things you would have thought to ask for. All of them are why the output does not read as amateur.

5. Review it before you publish it

Generation is half the loop. The other half is review_creative, which grades the result against four things at once — your brand rules, the platform, the intent you declared, and the content itself — and returns publish, revise, or regenerate.

It catches the specific failures that a props file cannot show you: a claim you cannot source, a pain point quoted in language your own brand rules forbid, a card that pauses too long on a still frame. If you are working through this alongside your first SaaS ad campaign, run the review before you spend a penny on distribution.

6. Attach the disclosure and post it

Every generated creative carries an AI content disclosure — a visible marker in the frame and a machine-readable manifest attached to the rendered file. There is no switch to turn it off, and that is intentional. See the platform's approach to labelling AI-generated content for what that means for you as the publisher.

Why trust this approach

Because the constraints are specific and checkable rather than asserted. The contrast target is 4.5:1 and it is measured, not assumed — set higher than the 3:1 the accessibility floor requires for large text, because social platforms re-encode video and low-contrast edges erode.

It is also honest about its limits. The platform never renders your video. That is not a feature we are hiding a roadmap behind — the render runs on your machine, which means you are the Remotion licensee. Remotion is free for a solo founder or a team of up to three, including commercial work, which covers most people reading this. At four or more people it requires a paid Company Licence, and holding it is your responsibility. We do not hold, provide or resell one.

That is a genuinely awkward thing to put in a blog post about our own feature. It goes here because you would rather find out now than after you render.

Where it will not help: if you have no positioning, no persona and no real call to action, the generator asks you for those instead of producing a video addressed to nobody. That is the same reason ad creatives fail without a designer — the missing piece is usually not the design, it is knowing who the thing is for.

Frequently asked questions

Do I need to know React? No. You run two commands. Knowing React means you can change things, which is a different benefit.

How long does the first render take? The first one downloads a headless browser and installs dependencies, so budget a few minutes. Every render after that is fast, and both are cached.

Can I use my own footage? Yes — you supply the URL and it is resolved on your machine at render time. A real photograph is the fastest way to stop a creative looking like an ad.

What if the video comes back wrong? Generate again. The system draws a different creative territory each time rather than returning the same cut with new words in it, so a second generation is a genuinely different concept — not a recolour.

Your next step

You now know the sequence: brief, objective, length, generate, review, disclose. The Mark Platform walks you through each one with the questions to answer and the tool that saves each answer, and it tells you what is missing before you waste a render on it.

If you want to see the full tool surface first, the MCP documentation lists what is available from your AI assistant. Otherwise — create your account and generate your first creative →