I knew it was too good to be true… Seriously though, big advancements, but far from perfect.
Our client liked the structure and format of the original animation (which was done in Keynote):
The one I generated was more abstract, but I asked it to be abstract.
Trying to recreate the above (without using a visual prompt, Veo 3 does not accept visual prompts), did not go so well. I tried asking Gemini to change the prompt, even giving Gemini a screen shot of the video to help it understand and prompt better, but it could not get the letters and numbers correct.
Here is the final prompt:
—————-
A 2D flat style animation, minimalist vector art, corporate explainer video style, 16:9 aspect ratio.
The entire animation uses a monochromatic color palette consisting of different shades of blue, from a very light sky blue to a deep navy blue. The background is a solid, light blue with a subtle, slow-moving animated pattern.
The video must be completely silent, with no dialogue or sound. The only text permitted is simple, clean alphanumeric characters (e.g., ‘A’, ‘B’, ‘C’, ‘1’, ‘2’, ‘3’, etc.) used specifically to label the headers of the table. No other on-screen graphics or text should be present.
The overall tone is clean, modern, and professional.
The animation will construct a data table with precisely aligned headers for six columns and six rows.
First, create the column headers: The capital letters ‘A’, ‘B’, ‘C’, ‘D’, ‘E’, and ‘F’ will animate into place horizontally along the top edge of the frame. Ensure each letter is centered directly above where its corresponding column will appear, with consistent spacing between each letter.
Next, create the row headers: The numbers ‘1’, ‘2’, ‘3’, ‘4’, ‘5’, and ‘6’ will animate into place vertically along the left edge of the frame. Ensure each number is centered directly to the left of where its corresponding row will appear, with consistent vertical spacing between each number.
After the letter column headers and number row headers are correctly positioned, a grid of clean, dark blue lines is drawn to form the six columns and six rows of the table. These lines should extend vertically down from each letter and horizontally across from each number, creating a clear 6×6 grid. The headers must remain outside of the data grid they are labeling.
Once the table structure with aligned headers is formed, small 2D shapes (circles and squares in various shades of blue) representing individual pieces of data will flow into the frame, populating some of the cells within the grid. Each shape should fit neatly within its cell.
Next, a stylized 2D icon of a magnifying glass, also in a shade of blue, will appear and briefly highlight a specific row by hovering to the left of its number header (e.g., ‘3’), causing the data in that row to briefly glow. Then, the magnifying glass will highlight a specific column by hovering above its letter header (e.g., ‘C’), causing the data in that column to briefly glow.
Finally, the shapes from a highlighted row will smoothly animate out of the table.
—————-
I then tried using Veo2 which allows you to add “Ingredients” in Flow as a visual reference. I used the same prompt above with the Keynote version as a visual queue. It took more than twice as long as Veo3. It was actually worse.
I then gave Weavy.AI a try. This is sort of like Perplexity for video – allowing me to try multiple models at the same time with the same text prompt and image reference.
It showed one thing, that Veo3 is better than the models I tried. 🙂 For some reason, Runway 4 didn’t accept my image prompt. Here are screen shots:
I then tried it without the image reference:
Testing also showing that Gemini generated video prompts work great in Veo3 and not so much in these other devices.
Still evolving….
