Yeah, its working for me now too...well, it did earlier so migration is nearly or fully completed I believe...

I have never ever shared these with anyone. But since you do need help with getting the prompt to give you what you actually want, I'm willing to share it with you but not in public. I refuse to share it in public because god who knows what would if I released it released...But, anyways, my method doesn't work 100% of the time but for the most part, it gives me what I want.

Ahhh, I see someone needs help with the image generator getting to understand the prompt...I could help with that...I have found found my ways for getting the image generator to understand the prompt the way I want it too but I am too afraid of saying it here...and there is a reason why I keep my methods/techniques to myself...but I can show you the results you want with my method (once the image generators get back up and running).

You were right all along! It is a Google model, more likely Gemma 4, or another Gemma model, or some other Google model idk lol like you mentioned. I used a few AI text generators as examples and both confirmed to be language models trained by Google! So know we know the AI text plugin is using a Google-trained AI text model. I wish we knew the exact model

My guy, this website is ran by ONE dude. Sure, he could hire more people, but this is obviously not that kind of website. This isn't some big corporate company, this is a community/user-driven website. You also have to consider the costs and resources. The dev may not have or make enough money to implement stricter safety features right now. He is paying for everything out of his own pocket. We are talking about paying more money just to add more safety features from a single developer. The best thing he could do is add a report button, at least for now. But when you are one person, running community/user-driven website for free, and allowing thousands of users to create and upload their own generators...Yeah, its not very feasible or effective for a website like this.

You may be right about it being Gemma 4! And it might be possible for Perchance to allow multiple AI text models, whether through the plugin or custom coding (idk lol). I have seen some text generators using supposudly using different models from apis but I don't know if its actually true or not. I suspect that https://perchance.org/ai-character-chat might not actually use Perchance's text model plugin. Based off of what you asked/instructed/prompted, it may be very well possible its Gemma 4, especially Google releases their AI text models as free and open source, so I'd least be a little surprised that's what the ai text plugin was using LOL

And I am even more convinced you might be correct because I found this on Google's website:

Earlier versions of generative models were only able to process 8,000 tokens at a time. Newer models pushed this further by accepting 32,000 or even 128,000 tokens. Gemini is the first model capable of accepting 1 million tokens.

https://ai.google.dev/gemini-api/docs/long-context

Which sounds similar too what your AI said

I found this in Elements tab from DevTools:

• "the text model is likely gemini model gemma4" (2d ago)

from https://perchance.org/ai-text-plugin

I didn't see it on the page, it was hidden. I don't know if it was from a comment or message (there was comment/message section) but apparently (if its an actual person who written it and not hidden text in code LOL) else believed it was also using Gemma 4. The other AI text generator I must've been using more likely used an external source (Huggingface and Deepseek, according to Network, seem like strong indicators)

Yeah, that's what I've noticed too. I heard something about the servers possibly in the process of being migrated to new servers. But the website is pretty unstable right now. I am not sure when everything will be back to normal again...sighs Hopefully, Perchance will be unstable yet usable for sometime until the migration is fully complete...

No, you are good! I understand that, it does seem that way tbh Not everything is free and open source. One software can be proprietary and free and the other can be open source and paid. True, a lot of AI models aren't open source, but we are lucky to have a few. I am not sure why I though DeepSeek, probably because it was free on the chat ui. Going through a whole reverse engineering would be complicated for me LOL. I think I knew they published like the weights and code but I was not aware the dataset was closed.

Yeah, me neither...the image generation isn't working either but the websites and the generators are working again...I'm assuming the outage will be fixed very soon. I can't generate anything right now...

Yes, but unfortunately, that its not the only type of AI image generator either…There’s some with the option of changing the age of your subject/character…I have seen a few of those type of generators…However, not all of these generators have the intention of wanting to produce CSAM/CP material. Some of them just want to include different ages/age groups for more variety and diversity. It can definitely be hard to tell whether the intentions are innocent or note; because you have different age groups containing underaged/minors (babies, toddlers, children, preteens, and teens) grouped in with adult age groups (young adult, mature/middle-aged adult, and elder/senior) and adult content for a wider range for character options or to produce CSAM/CP? If the adult content is extremely NSFW, then I agree that children shouldn’t be included at all. That’s borderline CSAM/CP.

In this case, it's been made very clear what the intention of this generator is. Preteen Girl and Teenage Girl included in gender-specific age groups is a huge red flag. Usually, with those types of generators, it's 18+...so I would strongly advise you to report that.

I think people seem to forget that this website is more of a passion project and community/user-based. Users are allowed to create, run, maintain, and moderate their own generators, rather than just multiple moderators utilizing a universal safety system. Trying to implement universal safety features for individual-ran generators probably wouldn't be very feasible right now. People seem to either forget or not know this website is run by one guy…He probably doesn’t have the money and resources to implement an enforced and stricter safety system and features. I don't think he even designed Perchance in mind like that and I mean it's not entirely his responsibility to oversee the safety and rule breaking of these generators; it's the creators job to do, they have the ability to do these themselves.

…Moderators, auto–moderation, auto-banned words/associations, report system, scanning/screening, wait time, pre-approval, requirements, background checks, task requirements, etc are not easy safety features to implement for one guy running. And there’s so many user-created generators, it would be hard trying to moderate and flag them down. On top of that, trying to stop people from bypassing it and there’s also people who generate this stuff for self-gratification and keeping it to themselves (by not posting it online) which can’t be done much with that…Now you can’t stop people from doing that. The best he could do is add a basic report button. I did see him mention it shouldn’t be possible to generate CSAM/CP if it's a bug in the image generator. I don’t really think it is a bug per say but a lack of censorship, more likely not censorship being implemented into the system, because if you are using an AI image model, then it can produce this stuff because it doesn’t come with built-in censorship. Censorship is associated with words used in the prompt that's deemed inappropriate. You have to add censorship as a feature. So it's not an actual bug, just a feature not implemented yet.

Now, it can be difficult to decide what’s inappropriate or not. As the dev mentioned, there is semi-nudity associated with children. Just nudity as a whole isn’t inherently sexual or bad in nature; many men, women, and children have been nude from dawn of time till now, semi nude, shirtless/topless, etc because there is a primitive/ancient, semi-primitive/ancient, or modern tribal/ethnic groups, culture/customs/traditions, environments, different cultures and places that normalize this in the modern or carry on custom and traditions, historical/ancient/prehistoric reasons, yada yada. This is a difficult grey area to figure out. And then there’s overly sensitive censorship which will completely blur or censor an innocent prompt.

@[email protected] Maybe you can add some type of AI chatbot thingy (safety analyzing chatbot), since AI is already so relevant, that analyzes the the whole prompt rather than just word-based? It will still analyze the words but will also analyze the intent and context of the prompt. If the user is typing an inappropriate prompt containing children with pornographic/sexual terms/words, then it will auto-reject the prompt. Or if there's a prompt written in a very suspicious way, it will request the user to rewrite, reedit, revise or auto-generate alternative prompts that's safer and more appropriate. I'm also thinking if the AI thinks something is inappropriate or suspicious but its not the user could chat with the AI clarifying their prompt and the AI will understand and allow the user to proceed with the prompt. However, the AI will reject anything that 18+/pornographic/sexual content associates children/teens/minors/underaged (you get the point) or other illegal things like zoophilia or animal sa. Just the concept of just chatting with the AI safety analyzing bot (what I'm calling it for now) like you would do with ChatGPT. However, the bot would have to be associated with the AI image generation textbox so it could analyze the prompt. It solves the problem with inappropriate prompts and overly sensitive censorship.

I think this would more likely be feasible for you with the ai chat text whatever plugin but consider how you would implement it universally.

UPDATE: Everything seems to be back to normal but the image generator is still not working properly...It keeps saying failed: fetch failure...

Thats a little bit too much high of a standard...devs can only do so much and people still find ways to bypass things. It doesn't matter if its a small, passionate indie project or high corporate service. The same thing applies to them both. The real problem is if they don't listen or don't do anything about it...

Yeah, I think something like that would be very useful and helpful. Its easier to use pre-made tools or features and integrating in your project or service. Consulting other projects is a smart thing to do because they can give you some insight about filters and censorship, stuff like. Filter for the Fediverse, like this one: https://github.com/db0/fedi-safety? Seems interesting, I also remember AI Horde, I think I used it a couple of times. It's great their filtering works as intended. I think the dev should take this advice, and maybe even hire or consult somebody to help him with this. I am not programmer or coder (I can't do this shiii without Microsoft's Copilot, it was the one helping with all the coding and troubleshooting lol. ) but maybe I can make a "filter" example, albeit a simple one, just as an example. I'd rather have filtered out words than Grok's horrible use of censorship, blurring images that's not even NSFW...

I kinda wanna aim for:

(child) ≠ [forbidden word]

or

(little\young\small) | girl/boy/child ≠ [forbidden word] Note: No specific order. Just the words contained in ( ) with (like girl/boy/child) to let the filter know to filter those forbidden words out those in combination of the other words.

(of course not the only one)

Filter system shouldn't censor innocent prompts, but rather specific ones based off the instructions its given. Child ≠ forbidden word example I've provided is a good way to add censorship without blurring images that uses the prompt: Two adults, Sally and Tom lays in bed together, wearing underwear. Sally wears bras and panties and Tom is shirtless in boxers. A prompt like this shouldn't be censored because Sally and Tom are adults, yet why is being censored anyways? If it allows NSFW as long as the adults are characters, then it shouldn't be an issue. This was Grok's mistake: allowing NSFW but had to enforce stricter censorship because the dumbahh developers allowed Twitter users to edited pictures in posts and creeps and pedos made sexualized images of women and children. Now Grok users can't generate any NSFW content involving consensual, legal adult characters only, after being marketed as "unrestricted/uncensored" type AI image generator. This is the type of issue we want to avoid. Perchance allows NSFW and not have this "family friendly" or "all ages" image. So allow NSFW content for any prompt containing adult-aged characters only. Shouldn't be too hard...

I generated these with my prompt. I think this is what you were going for. Some of them were "cherry picked" as my way of prompting doesn't always work, but these were absolutely generated with my prompt, many of them actually (when I increased the numbers of images). I used some of the AI image model's (a random generator from Perchance) settings like guidance and art style (didn't really help much until I put real human being/photograph in the prompt...bruh moment). No flyaways, no buns :3

I can share a erm vanilla version of the prompt, should get the about the same result:

woman with long and tall ponytail that sits up high on top of head, the ponytail is pulled-back-tightly/slicked back with gel and stretches the forehead. The woman is a real human being. She is in a professional photograph. Avoid styles, increase guidance (try 12-15), and increase number of images (10 is okay, 20 is good, 30 is great, 40-60 is amazing!)

Ah, okay. I see horde-safety kinda like integrates the Fedi code in their filters.

Fedi-/Horde-Safety use an image interrogator, to query what’s actually in the image. That’s immune to that. They also have all these heuristics. (You can have a look in their code.) It’ll come with false-positives, though. It’s a trade-off.

Huh, now that's pretty interesting. I will defiantly take a look into their code, I want to see how they programmed it.

Mind you’re currently in the process of fleshing out your own interpretation of the wheel. My recommendation was not to do that. And your approach has the usual downsides: People can use other words, synonyms. Describe it. Or leverage wrong spellings. Jam an adjective in-between. Do an exact request, but in Chinese language… All that will circumvent filtering by simple word filters. And that’s usually what happens 5 minutes after you roll out such a system. It’s a) pretty much not effective at all. And b) it’s complicated and you’ll end up in a cat and mouse game.

Yeeaaaahhhh...I just realized that that now...emphasis on the process of fleshing out my own interpretation. I wouldn't recommend as something to actually make and integrate as censorship. I think I was literally doing that, trying to showcase how I thought how the censorship/filtering worked, like the basics of it...I knew it wouldn't be as accurate or effective because not all censorship/filters actually work like that...I guess in my head it was more of a "demonstration" rather than an actual tool/solution. I think I meant to demonstrate how the censorship/filter on what words and associations to focus on instead of banning innocent prompts because of too much censorship (Unfortunatly, that what Grok does. Because they effed up knowing releasing the ability someone's post without consent and now it all backfired, cranking up the censorship and making it practically unusable. And the fact the government is enforcing it also makes it worse) ...But you are 100% right: its not effective since people can use ways to circumvent the filter like you said, use different words, synonyms, mispellings, sentences, etc. (Man, I wouldn't be too surprised if it was less than 5 minutes...its insane...). Very ineffective and complicated...I just thought about its trying to keep adding words when people can bypass is hard as hell. And the cat and mouse game? God, couldn't be more true. Trying to add more words to the filter and people STILL find ways to bypass would be exhausting.

I think only that AI text generator was using DeepSeek through huggingface api, not the actual ai text plugin. I didn't mean to mistake DeepSeek as the default text model for Perchance, I agree with you, it does seem to be more likely Gemma for sure.

I actually found out through using DevTools to extract/analyze the data! Using network, I found these:

1. https://huggingface.co/deepseek-ai/DeepSeek-R1-0528/resolve/main/tokenizer.json

2. https://huggingface.co/api/resolve-cache/models/deepseek-ai/DeepSeek-R1-0528/4236a6af538feda4548eca9ab308586007567f52/tokenizer.json?%2Fdeepseek-ai%2FDeepSeek-R1-0528%2Fresolve%2Fmain%2Ftokenizer.json=&etag=%22d81c3a6e2093a07ae6fa596d9b154f6669aafafd%22

3. https://huggingface.co/deepseek-ai/DeepSeek-R1-0528/resolve/main/tokenizer_config.json

4. https://huggingface.co/api/resolve-cache/models/deepseek-ai/DeepSeek-R1-0528/4236a6af538feda4548eca9ab308586007567f52/tokenizer_config.json?%2Fdeepseek-ai%2FDeepSeek-R1-0528%2Fresolve%2Fmain%2Ftokenizer_config.json=&etag=%22c24fd8eb5753a737488f8e447cb5739b0e9f7a1a%22

You're right, there was no DeepSeek model api, I think I meant to say the huggingface api.

Only from https://perchance.org/ai-character-chat though. I haven't looked at other AI text generators/tools.

And the fact it said the cut off date was Jan 2025 and it kept insisting it was Google makes me believe Perchance is actually using Gemma 4 or other Gemma models.