Moderation system improvements
Violent content that is descriptive (e.g. historical in nature) is being blocked
“““
Hi, first of all, Your tech is amazing.
I’m a pro subscriber and recently ran into a censorship issue. I create video-based learning objects (among other things). I’m currently producing a series of side-by-side comparisons of pre-colonial warrior cultures in which I have representative avatars speak to how they did things and why. I was describing how and why Malay sea raiders took women as slave breeders. Your system blocked me. Is there a challenge mechanism in the same midjourney does? This’ll be a hindrance to me; I am after all teaching about violent cultures.
“““
Log in to comment and vote
Comments8
Michael Lingelbach
Mar 10, 2025
We’ve adjusted this down, and we think we’ve hit a nice balance
Dan Merrill
Dec 10, 2024
Seconding the too-strict celebrity moderation. My own selfies are often moderated and alas, I am not a celebrity. The same image sometimes passes and sometimes fails, it does not seem to be deterministic. I hope to provide fun selfie lipsyncs via the Hedra API, but if my end users are falsely moderated I cannot provide a usable platform.
Michael Lingelbach
Dec 21, 2024
This should be fixed
Michael Lingelbach
Nov 19, 2024
I’m going to merge this with the other moderation option, but we do have a big improvement in moderation coming next week. An adult character in only underwear will likely not trigger the new system.
Michael Lingelbach
Nov 9, 2024
Unfortunately we use a third party service for this and have limited ability to improve it other than reporting to them.
John
Oct 10, 2024
I wouldn’t hold your breath. I was recently censored on HeyGen because my avatar reported a news story about a man running, along with his personal AI bot, for mayor of Cheyenne, WY. They “don’t allow politics.”
This sickening “woke” environment makes it more difficult for uncrazy people to practice free speech. I suspect Hedra and the others are terrified of the government hitting them with political lawsuits.
Michael Lingelbach
Sep 25, 2024
It usually helps to reword the language to emphasize the factuality rather than the violence, but I understand the conflict here. Perhaps this will be better with the next version of our moderation system.
Michael Lingelbach
Aug 27, 2024
The filter system is not easily tuned as it’s provided by a third party vendor unfortunately.