Effectively tell a customer story in 1 min, using 3 AI tools
Tools mentioned: ChatGPT, Playground.com and Invideo.io.
A recent article, Create a product vision video using AI, for free, in less than 1 hr, became one of the most read pieces in my little blog. Since I enjoy both e̶x̶p̶l̶o̶i̶t̶i̶n̶g̶ leveraging AI to do my job and sharing my learnings, I thought I’d keep it rolling.
I may speak for quite a few Product Managers here. We run many webinars, for customers, internal functions/stakeholders, marketing endeavours, etc. Those PowerPoints or Miro boards could make us feel tiresome very quickly. And they don’t seem to deeply engage our audience either. Thus I’d try different presentation tools and formats whenever I get a chance. That was when I discovered that I could actually wrap up a vivid product vision or problem statement in a 1–2 minute video without having to do much work.
Steps:
- Start by writing the video script.
- Divide the script into individual scenes.
- Then generate one image per scene.
- Finally, generate a video using the script and images.
Start by writing the video script
Before introducing any AI tool that writes the script, it’s necessary to say that it’s the purpose and goal of the webinar that drives the script, not the AI.
In my case, the goal was to communicate a new product feature to internal functions, i.e. Sales, CS, and Consulting. The product feature, in a tiny nutshell, gives blue-collar workers the ability to quickly and easily read safety documents, so that they know how to protect themselves. And the reason we developed this feature can be told via a story:
A bustling construction site. Many safety hazards. Workers who are vulnerable to machineries, electricity, and ergonomic risks. The pressure is high. The place is loud. Safety procedure must be followed to prevent incidents. And workers must always have access to their safety instructions.
That’s basically the idea for the script. Now just ask ChatGPT to enrich it. And don’t forget to help ChatGPT understand the task in detail, e.g. how long should the script be.
Divide the script into individual scenes.
Divide the script into individual scenes based on logical sense. I asked ChatGPT to do so first, but decided that my own idea was more suitable. (Human 1: O AI)
I needed the video to be no longer than 1 minute, cuz I’m running a live webinar after all. A good intro via an AI video could catch attention and set the stage. But playing a video too long would harm the engagement. This means I had 5 images to generate. One for each scene.
Then generate one image per scene.
I chose Playground.com for the quality and speed, but the tricks below work on other tools as well, e.g. Midjourney.
There’s no need for a user guide, as the software is rather intuitive, with common patterns:
- Prompt to image
- Adjust image filters
- Choose between different tech models, etc
It’s very straightforward to generate images one by one. But soon you’ll hit a wall — the image styles are disconnected. You can’t put them in the same video, where one image looks ultra-realistic and colourful, whereas the next one is completely the opposite. For putting together a video, it’s better if the images are of the same series.
Use the Seed number to create a sequence of images.
Navigate to the “Seed” pane of an image that you like. The Seed number carries and maintain certain characters in a generated image.
Copy the seed number. Paste it to every following image that you are about to generate. This will ensure that the images have a consistent style, like paintings from the same artist. The same works in Midjourney too.
Use Image to Image for similar image layout and colour
If you would like AI to generate more images based on an existing design, upload your design here. The generated image will look incredibly similar to your design.
Note that the AI may prioritise mimicking the look rather than your prompt or logical sense.
Manually balance the gender gap
It came as no surprise that the word “worker” resulted in male workers in every generated image, and every “female worker” resulted in heavy make-up and strongly suggestive female body characters. To manually balance it, try Exclude from Image.
I attempted to use it to remove the heavy make-up. It hasn’t worked too well. If anyone could share a tip or two, I’d highly appreciate it.
Finally, generate a video using the script and images.
I used invideo.io cuz it covers the audio narrative as well. As I mentioned in the previous article, their AI voice quality is beyond satisfactory. On top of that, Invideo can automatically refine the video script if you want, and add subtitles to the video. For a short story, this is all I need.
I liked the interactive dialogue way of making adjustments to the video: you just type your demand.
In the end, working on the script, the images, and the video took me 1–2 hours. How long would that have taken me if I was drafting animations in PowerPoint? It would’ve been shorter if I'm being honest.
However, the ability to work with AI tools can only be more and more impactful, and the webinar vibe was much more lively than walking through a good ol’ PPT.
