Veo 3 (probably) solves the character consistency problem
Veo3 may have cracked the character consistency nut. Find out how
I documented how current image generation tools weren’t helpful when creating character sheets, but Veo 3 (accessed through Google Gemini) may have solved this problem.
It’s not perfect — it still adds a background even when asked not to, you still need to output the video file, cut it into frames, and then pick your frames — but the video itself is consistently rendered without artefacts and the UI is much more straightforward to use.
Here is some example output when I was rendering our friend Chad from Hot Crossed Lines. In the prompt, I asked Gemini to show him “full body, in a neutral stance, and then have him become angry, then laughing, then sad.”
It wasn’t too hard to generate these stills. With basic post-production they can get transparent backgrounds and be perfectly usable in other applications.
I experimented with more expressive body movements, and generated this rather fun variation. Note how the sloth mimics the Chad’s expression even more closely. It just did this and it’s charming.
The final prompt shows the limits — and possibilities — of this technology. I asked “output a single frame of the character that best captures each of their different emotions” to see if I could skip the post-production steps, and it completely misunderstood my prompt. Instead, it produced this, which I will share without comment.