MiniMax H3 for Founders Making Their Own Ads

Founders making their own ads face a specific problem: no crew, no budget for one, no appetite for learning an editing suite, and the marketing is competing with everything else on the list. Generated video removes the production step, which leaves you with the part that was always yours anyway, deciding what to say.

What should a founder actually make?

Not a product tour. Not a brand film. Five different hooks for your single clearest use case.

This is the most common mistake and it is worth being blunt about. Founders reach for the comprehensive explainer, because they know the product deeply and want to convey all of it. That video reliably underperforms, because nobody has agreed to care yet. The job of the first video is not to explain, it is to make someone want the explanation.

So: pick the one use case you are most confident about, and write five different promises against it.

Where do the hooks come from?

Your support inbox and your sales calls, not your imagination.

The sentences that work are the ones customers already say. "I was doing this in a spreadsheet for two years." "I did not realise it was costing us anything." "We tried to build this internally first." Those are hooks because they are specific and true, and someone in your market recognises themselves in them.

What does not work is your value proposition read aloud. "The all in one platform for growing teams" is a positioning statement, not a reason to stop scrolling.

Write five. You are not choosing between them yet.

Do you have to be on camera?

No, and for most founders that is the unlock.

With no reference images attached, MiniMax H3 runs as text to video, so it generates the person as well as the scene. You write the line, the model produces someone saying it, with the mouth matching because sound is generated with the picture.

A man in his early thirties sits at a tidy desk in a home office, turning from a laptop to face the camera. He says, "I was doing this in a spreadsheet for two years." Medium close up on a 50mm lens, static camera at eye level, soft daylight from a window on the left, clean natural colour. Quiet room tone, faint keyboard sounds, no music. Vertical framing.

Change only the spoken line for the other four. That constraint is what makes the batch a test rather than five unrelated videos.

If you do want to be on camera, that is a different and also good route, and our note on ads for camera shy founders covers the middle ground.

How do you show software?

Software is harder than physical products, because the interesting part is on a screen and the model will not render your interface.

Two approaches that work:

The reaction, not the screen. Generate the person having the realisation, and show the actual product in a screen recording you cut in afterwards. The generated clip carries the emotion, your recording carries the proof.

The phone in hand. A person holding a device toward camera reads as a demo without needing the interface to be legible:

A person in their late twenties holds a smartphone toward the camera, showing an app on screen, and taps through two features while speaking. They say, "It took me longer to sign up than to make the first one." Medium shot showing their upper body and the phone, screen clearly visible and free of glare, modern minimal background, even soft lighting. Natural voice, subtle interface tap sounds, no music. Vertical framing.

Do not ask the model to render your actual interface, your logo or any on screen text. Lettering is unreliable across every model in this class. Cut real screen recordings in during the edit, which is more convincing anyway.

What does a weekly routine look like?

One session, repeatable:

1. Pull three sentences from support conversations this week.
2. Turn each into a hook line, plus two you invent.
3. Generate all five at five seconds and 480p, vertical, changing only the line.
4. Watch them muted on a phone. Keep two.
5. Rerun those at 768p, add captions and an end card, launch both.

The generating is the short part. The thinking is the long part, and that part was always yours. What changed is that it is no longer followed by a week of production.

The full routine is in how to draft a video ad in minutes.

What is the most common founder mistake?

Making the video you want to watch rather than the video your buyer needs to see.

Founders know their product better than anyone, and that knowledge is a liability in a five second clip. It produces videos that begin at step four, assuming the viewer already agrees the problem is worth solving. The viewer does not, because they have not been thinking about it for two years.

The fix is uncomfortable but reliable: start where the customer is, not where the product is. The first line should describe their situation, not your solution. "I was doing this in a spreadsheet for two years" starts in their world. "Our platform automates workflow orchestration" starts in yours.

A related version of the same mistake is length. Founders write long because there is a lot to say. But a longer video does not get more attention, it gets less, because the decision to keep watching is made in the first second regardless of how much you queued up behind it.

Write the shortest version that makes someone curious, and let the landing page do the explaining.

What about launches and announcements?

Same method, different content. A feature announcement is a hook plus a demonstration plus a call to action, and each of those is a short clip you can generate and assemble.

Our guide to SaaS feature announcement videos covers the structure. The thing to resist is making the announcement longer because the feature feels important to you. Length is not respect, it is a tax on the viewer.

What should you do with the results?

The output of a test round is two ads and one piece of knowledge, and the knowledge is worth more.

When one hook clearly beats the others, you have learned which promise your market responds to. That answer does not stay in your ad account. It belongs on your landing page above the fold, in your cold email subject lines, in your onboarding copy and in how you describe the company when someone asks. Most founders discover their real positioning this way rather than in a workshop.

Write the winning line down somewhere permanent, along with what it beat. A record of which promises lost is nearly as useful, because it stops you rediscovering the same dead end in six months.

Then run it again next month with five new guesses. Markets move, and the hook that worked in spring is not automatically the one that works in autumn. Platform guidance moves too, so it is worth periodically checking the current creative specs from Meta and TikTok for Business.

Is this good enough for a real company?

For paid social and organic short form, yes, once captions and a proper end card are on it. The viewing conditions, a phone at arm's length with a thumb ready, are forgiving of render quality and unforgiving of a weak first second.

For a keynote film, a homepage hero that runs for a year, or anything where a viewer is genuinely paying attention on a large screen, use a slower model and take more care. The distinction is laid out in drafts against final cuts.

Being honest about which one you are making is more useful than any tool choice.

Frequently Asked Questions

Do I have to appear on camera myself?

No. Text to video generates a person delivering your line, so you can produce talking head creative without filming yourself or hiring anyone.

How long does a founder need to spend on this per week?

One focused session produces a batch worth testing. The thinking about what to say takes longer than the generating, and that part is yours regardless.

What should a founder make first?

Five different hooks for your single clearest use case. Not a product tour, not a brand film. Find out which promise lands before you build anything longer.

Is generated video good enough for a real product launch?

For paid social and organic short form, yes, once you add captions and a proper end card. For a keynote film, use a slower model and more care.

Should the person in my ads be consistent?

Within a campaign, yes, using reference images. If you want a recurring spokesperson across everything, the actor tools are more reliable than raw references.

What if I do not know what my hook should be?

That is what the test is for. Write five guesses drawn from your support inbox and let the results tell you, rather than deciding in advance.

Start this week

Pull three sentences from your support inbox, write two more, and generate five clips before you decide anything. Plans start at $29 a month with credits included, detailed on the pricing page. If you want the prompt craft first, the prompt guide breaks a brief down line by line.