Vibe Coding and Vibe Graphic Design

I had always watched the vibe coding debates from the side line.

As someone who knows coding and has used vibe coding tools, vibe coding always feels both like a blessing and a curse to me, but I cannot exactly pinpoint the reasons.

That is until I had to use generative AI in a similar yet different way - i.e. to create digital videos without any digital graphic design knowledge.

By using a mix of (paid) tools, I was able to get something that somehow resembles what I wanted - i.e. a video to describe what a team of data scientists does at work.

(For those who are interested, I used Gemini to first create a base image that represents the style that I needs. Then I used Hailuo agentic workflow to create the storyboard, scenes, video clips, and narrations.)

This is an amazing achievement for someone like me who can only draw stick figures. But as the AI-generated video gets longer, it is easier to spot small inconsistencies in between frames.

Correcting for those frames using prompts would be possible. But that would mean having to regenerate the videos multiple times, where costs can add up very quickly.

It can also be a very frustrating experience. Because you may be able to get something you need in 2-3 prompts this time, but the next time you may not get what you need even after 20-30 prompts.

In the end, I resorted to using the AI-generated video as a base. And then applied my limited video editing knowledge to remove inconsistent frames and add background music.

In many ways, this reflects the ways of working with vibe coding tools as well - it is important to use the tools to bridge your knowledge gap, although the quality of your final product is always limited by your existing knowledge.