Quote from: EssentialDarkNote on Yesterday at 12:10:05 am

Is there any chance you want to share a link to your content to showcase?

I'm hoping to start a thread here soon! I have many hundreds of pictures, but I'd have to comb through them to find some good 'standalone' images that suggest a self contained story

Quote from: EssentialDarkNote on October 06, 2026, 02:32:51 am

I would point out that GPT is a lot less prudish than it used to be, and as long as you tell it all the sex scenes are 'adults making out', its willing to let you take it in some pretty spicy directions (provided you are willing to edit and control the flow paragraphs at a time, and write the spicy scenes yourself or with another model).

I have noticed in general AI seems to be getting a bit better at allowing more sexual/suggestive material - hopefully this is a sign that a happy medium is slowly being reached between "completely bland & sexless, never spicier than a workplace motivational poster from the 1990s" and "you can go ahead & do anything, even the most depraved illegal material is fine, surely there won't be any blowback from this".

Update: it seems that within the last 24h Google Flow has updated to Nano Banana 2.1 (and it might be my imagination, but I feel like the usage limit is much higher - although this might just be temporary, or perhaps there was a mid-day reset when they did the update). I haven't played around with it too much but it seems subtly better at getting the proportions between characters & objects right, and at replicating various film styles. The big improvement I notice is that it seems way better at posing characters naturally in the environment in a wider variety of ways. In the past there was a tendency to default to the old "character standing in a neutral pose in front of a background that they don't interact with, which may as well just be a painted wall". Now they're much more likely to be interacting with nearby objects in organic ways, and the lighting on characters feels much more cohesive with the surrounding environment.

The other change I've noticed (which, again, might just be my imagination) is that the new model seems less likely to reject prompts, choosing instead to simply ignore the most offensive elements. So, the pro here is that you'll get more images generated (rather than just wasting tokens), and the con is that perhaps you'll have less success occassionally tricking the AI into giving you something truly spicey. In general, in terms of what it'll allow, it seems comfortably PG-13, although if one were to start leaning very heavily into certain kinds of fetishistic language it might push back on it a bit.

Looks like all the suggestions so far have been for images, video, and audio and not really anything for text, so I'll give my recommendations for a local LLM that can help with text gen, and roleplay.

First, get https://github.com/thomas9120/LLama-GUI and install it. This is what allows you to run and interface with the LLM.
Then get the LLM. I recommend https://huggingface.co/DavidAU/Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-NEO-MAX-MTP-GGUF if you have less than a 5090 to run it. (lower than IQ4_XS quantization is not recommended)
Next, get https://github.com/SillyTavern/SillyTavern for the primary interface. It is basically roleplay chat, and you can set up various different characters, world rules, environments, and all sorts of things. It also seems to help bypass the few remaining censorship refusals that the LLM model I recommended has through basic jailbreaking techniques. (you can download several characters and lorebooks here https://character-tavern.com/ )

I highly recommend generating no more than 40 paragraphs at a time, and refining your prompt each time to get the output you actually want, or close enough to allow manual editing. (though if your prompt is long and detailed enough, you can produce as much as 100 paragraphs in a single go)