Upgraded to Google AI Ultra and the false policy flags got worse, not better anyone cracked this?
Posting this hoping someone here has figured it out, because I've spent more time fighting the filter than actually making anything.
Quick context:
I run a small AI film channel and build everything around original characters no real people, no celebrity likenesses, no suggestive content.
I upgraded to AI Ultra ($99.99/mo) partly for more credits to generate since I'm working on building a whole cinematic universe around my channel name branding " AI Night Fall Gallery" and also was hopping for the better output and partly hoping the moderation would be less twitchy. It's been the opposite.
Almost every generation comes back with "this prompt might violate our policies about generating prominent people" for prompts that are nothing more than a fictional character sitting in a studio talking to camera. I've also gotten generic "this generation might violate our policies" with zero indication of what tripped it.
But I also observed that when I prompt for output that to should help generate AI slops like some of my TickTock or IG reels then it's all good but when I tried to work on project that seem like I trying to do this for real thats's when those filters start to popping up.
What I've already tried:
- Rewriting prompts in purely technical/cinematic language (shot type, lighting terms, lens no character descriptors that could read as a real person)
- Dropping any brand names or third-party references
- Switching models Veo 3.1 Lite or Veo Fast, Fast, etc. Same flags across all of them
- Regenerating identical prompts that succeeded earlier in the same session (sometimes works, sometimes doesn't, which makes it feel random)
The "prominent people" flag is the confusing one these are characters I designed from scratch using Nano Banana . There's no real person for it to match against.
So, genuinely asking the people who do this daily:
- Is the "prominent people" flag actually matching faces, or is it keyword-triggered?
- Has anyone narrowed down which words set it off?
- Does locking a character to a reference image reduce the flagging, or make no difference?
- Has anyone found the moderation behaves differently across tiers, or is Ultra the same filter as everyone else?
Screenshots that include some of the chats attached so you can see I'm not hiding some edgy prompt. Any real world fixes appreciated.