pictv

I've been building PicTV, an AI image-to-video product, and one problem became obvious very quickly:

Most users don't know how to write a good video-generation prompt — and they shouldn't have to.

A normal user might upload one or two images and say something simple like:

“Turn these into a 12-second UGC product video. Make it feel natural and have the creator introduce the product.”

But getting a video model to produce that reliably can require a much more detailed instruction:

• who the person is

• which image represents the product

• what happens in each time segment

• camera framing

• product consistency

• dialogue

• audio

• movement

• what must not change

That can easily turn into hundreds of words.

The more I worked on this, the more I realized that exposing all of that complexity to the user is the wrong product design.

The user should provide the idea, not the production specification.

So the direction I'm working toward with PicTV is adding an AI “director” layer between the user and the video model.

The workflow becomes:

Images + a simple human instruction → AI understands the intent → builds the scene plan and detailed prompt → video model generates the result.

For example, we recently tested a workflow using two reference images:

• one image of a creator

• one image of a small handheld fan

The system then generated a horizontal UGC-style product video where the creator introduces the problem, takes out the fan, turns it on, reacts to the airflow, and recommends it.

What interests me most isn't just better prompt engineering. It's removing prompt engineering from the user's job entirely.

I think AI creative tools will become much easier to use when the interface stops asking users to communicate like model engineers and starts letting them communicate like normal people.

That's what I'm building with PicTV.

PicTV: [YOUR_PICTV_URL]

I'd be interested to hear how other founders building AI creative tools are handling this. Do you expose advanced prompting to users, or hide most of it behind an intelligent layer?