enthusiasm_headquarters

Yeah famous people were removed. Monroe is so iconic that you probably can't isolate the training data, the same for most presidents or elvis or people like that.

Try throwing some minor variable data into the descriptions of your character, example: A casual photo of a genuine {23|24|25|26|27}-year old woman, she has an {aquiline|up-turned|celestial|button} nose, and is of {German|Dutch|Belgian|French|Welsh} ancestry.

The actual characteristics don't mean anything in real life, but you can get it to stop being so monotonous in the renders this way. Sometimes a simple change of age changes the template of whatever its rendering.

Another thing to try is t not rely on presets. Professional Photo is especially annoying because of how limited the "look" is. Instead, try "No Style", and hack up the presets. Try taking some of the added tags from 90s photo for example.

Also, telling it to be unique sometimes actually helps. "She has a unique beauty" and so on.

It's really annoying but possible to get it to stop doing what you describe.

The AI has improved in its ability to create two different entities with unique attributes, though the tendency will still be to merge them. If you tell the image model a name, say, Samantha, it will use training data about Samanthas to influence the outcome. It doesn't matter if you are declaring Samantha to be this or that.

The best you can hope for is something like: A casual photo of two different people. One is a woman with dark brown hair, aged 25, wearing a white jumpsuit and laughing. The other person is a man, aged 27, who is wearing a red jumpsuit, but he looks very upset.

You'll notice that it mixes the two characters up, because it doesn't understand the human understanding of these values. One thing I've noticed is that putting a lot of gender pronouns in a prompt increases the "gendery-ness" of a render. Just putting "she" and "her" several times increases the chances you'll get two females, even if you only describe one. The AI doesn't remember things in any coherent sense.

Ohh. I hadn't even tried going to the link. I guess try contacting the user that made it? If they weren't anon...

It looks really sharp and a lot has been put into it, I'm surprised there is no contact info anywhere or even a name.

Dude what fucking planet are you living on?

If you're asking if the AI understands persistent narrative threads across characters in a chat, nope. You can do some tricks like write entries in the lore file of the main AI character you are talking to that remind it about the guest characters in the session, but the AI will tend to favor the main AI character in all things.

It's possible to get characters to act very differently, but not without a lot of work. All the guest AI NPCs invited will slowly morph into the main character. This tool is just not quite ready for any kind of narrative depth. It's challenging enough to get something solid with one AI.

I was able to save a .json file to my Downloads directory and load it up again just fine. Assuming you meant: https://perchance.org/ai-story-generator

4/6 outputs worked. You probably have to drill down into your prompts and figure out what is causing the problem. Juggle the priority of different terms. Those closest to the top of the list are emphasized by the AI. You can get drastically different renders by just rearranging prompts.

Also, neg prompts don't work the same as they did. They work, but not quite the same... I can't explain it any differently.

If your primary reason for using Linux is to play all the newest games, don't use Linux. You cannot expect open source to compete with a game software monopoly. They intentionally have it this way. That being said, you can use Proton and Wine to play most games, but again, if you absolutely must play the newest releases and that's your reason for going Linux, don't main Linux.

For a few years what I did was just dual boot Mint with Win 10. I did most things on Linux as I tried to learn it, but would boot back to Win for certain work reasons or to play the newest game. Now that I don't care about new games so much, I erased my dual boot partition and am 100% Linux. I play many games on Steam, which is made on Linux, and I'm just fine.

Use Wine and Proton instead,

It pleases me to read these things. I worked for M$ and coupled with more intense reasons I left Win/OSX about 10 years ago, and have never looked back. Carry the flame.

I noticed you outright block vpns in the chatrooms. That certainly helps. Systems I've seen that rely on voting can certainly be abused too, of course. I am happy to hear that there are already volunteer mods on here, and yes I was speaking from experience in the community management world, not as an author.

Thank you for your great project. I didn't mean to come off as overly pessimistic. Some scars... do not heal...

I was a volunteer moderator on a pretty popular game community for about 6 years, where the active userbase sometimes exceeded 10,000 online at any one time. We had plenty of 8ch garbage people, coordinated spamming from nazi IRC channels, Trump assholes just being Trump assholes, and sketchy CP pervs skulking around.

I am as annoyed by moderation as most people are, but people aren't going to moderate their own chatrooms and communities reliably, if at all.

If you're thinking of a scripted or AI moderation system, good luck, though who knows. Maybe AI moderators might work in the near future, though they'd be pretty fun targets to troll. Word bans, blacklists and offensive syntax filters are a nightmare to maintain, that's the way channels have been moderated for the past 30 years.

The sad truth is that safe places require human moderators who care about the community to spend time protecting it. It's a job nobody wants to do and burnout is a bitch. If there were a few trustworthy people who have a lot of history with perchance and the community who could be online, having it set up so that users could just ping them reports of abuse/scams/CP and they would have the power to intervene on anyone's thing that was opened to the public.

That's the only way. Spending time trying to code a way around this is going to suck your brains out.