What AI models actually allow. Tested, not assumed.

2025 and 2026 turned AI adult content from a legal grey zone into a regulated space with real enforcement: a US federal crime for non-consensual imagery, deepfake laws in 48 states, UK fines for missing age checks, EU labeling duties. Here is the map — and the one principle that organises all of it.

Everyone can tell when an AI image is off, few can say why. Each artifact has a specific technical cause and a specific fix — hands, limbs, faces, skin, teeth, text, seams, smear, flicker. This is the field guide, for images and video.

"Treating adults like adults" was the promise in October 2025. Ten months later the adult mode is paused with no timeline, ChatGPT still refuses explicit content, and the search interest hasn't gone anywhere. Here is the full story, the current policy line in practice, and what the alternatives actually are.

AI sexting went from novelty to mainstream in two years, and most explanations of it are either marketing or moral panic. This is the technical version: which models power it, what happens to your messages, how voice notes are generated, and the honest checklist for judging a platform.

LoRA, checkpoint, abliterated, i2v, score tags, vendor filter — the adult AI world runs on jargon, and half the arguments online are people using the same word for different things. One page, every term, in plain language, linked to the deep dives.

Lightricks put a 22B video model on Hugging Face in August with a permissive licence, day-one ComfyUI support and no content filter. In a month where Alibaba and ByteDance shipped stronger but closed models, this is the release that changes what adult creators can run themselves.

Alibaba launched Wan 3.0 on 24 August 2026 after a public beta since 6 August: 30-second single-pass clips from text, images, audio, video and now documents. Like 2.5, 2.6 and 2.7 before it, it is hosted only. Here is what shipped, what the hosted model does in practice, and what it means for adult creators.

Seedance 2.5 doubles the maximum clip length and tightens consistency over long shots, but it is not a straight upgrade for adult work. We compare it with 2.0, 2.0 Fast and Mini on the same prompt set, cost per usable second, and where each tier breaks.

Seedream 5.0 Pro is the best closed image model ByteDance has shipped and a lot of sites claim to offer it uncensored. The model is API-only and moderated; the uncensored offers are older or different models. Here is what the official model does, what the claims are worth, and what to use for adult images instead.

"Stable Diffusion porn" is still the most-searched phrase in adult AI imagery, and the answer has moved on from the 1.5 days. Three SDXL-derived families do the real work now. We compare them checkpoint by checkpoint, explain the prompting differences, and lay out what running them actually costs.

Wan 2.7 dominates the NSFW video conversation, and most of what is written about it is wrong: the weights were never released. We explain what API-only means for adult use, why the last open Wan is 2.2, where 2.7 runs with adult content enabled, and when self-hosting still makes sense.

Qwen3.8-27B landed with an Apache 2.0 licence, vision input and a 262K context, and the "uncensored" versions were on Hugging Face the next day. Most of them are abliterations, not fine-tunes. Here is what that distinction means for adult roleplay, and when a real successor to the 3.5 writer tunes is likely.

Grok Imagine earned a reputation as the closed image model that lets you get away with more. We tested exactly how far that goes through the x.ai API: lingerie, implied nudity, explicit prompts, and celebrity likeness. The answer is more permissive than OpenAI and Google, and still a hard no on explicit content.

Meta is back to open weights with Muse Glimmer, a 30B model under Apache 2.0 that runs on one consumer GPU. It was built for tool use, not conversation, and it ships safety-tuned. Whether it ever matters for adult roleplay depends entirely on the fine-tuning community, so here is what they have to work with.

Seven video models, one adult prompt set, one question: which ones can actually make NSFW clips, and how good are they once the filter is out of the way? A ranked comparison with the reasoning behind every placement.

Wan 2.7 is the better video model on every quality axis, yet Wan 2.2 is still in production on adult platforms. The reason is weights: 2.2 is the last open Wan, so it is the only one with an act, pose and camera LoRA ecosystem. This guide explains the trade-off and when each version is the right call.

Pony Diffusion XL is the base under more adult image checkpoints than any other model, and most people prompt it wrong. This guide covers the score-tag system, the difference between anime Pony and realism merges, the LoRA ecosystem, and the prompt structure that gets consistent explicit results.

Flux is the most capable open image family since SDXL, and the most misunderstood when it comes to adult content. The base model is not trained for it, hosted versions filter it, and the community fine-tunes are where the real capability lives. Here is how the pieces fit.

Illustrious took the anime crown from Pony in 2025 and its realism merges are now credible for photoreal adult work. This guide explains what it does differently, how to prompt it, where its character knowledge helps and where it is a liability, and how the realism merges compare with Pony and SDXL.

Google's Gemma 4 ships safety-tuned and refuses adult roleplay. The community fine-tunes built on it do not, and the best of them are the most natural persona chat models available at their size. We explain what a fine-tune actually changes, why the 26B class is the sweet spot, how "for her" and "for him" tunes differ, and the failure modes to watch.

GPT Image 2 is the most-used image model in the world and one of the most-searched for "NSFW". It cannot do it, and the reason is structural: OpenAI trained the refusal into the model rather than bolting a filter on. We explain what that means, what the API's moderation parameter really does, and where the line sits for suggestive content.

Kling is one of the most-searched models for adult video and one of the least usable for it. We mapped exactly where its three filter layers trigger, why the "uncensored Kling" sites are not what they claim, and which models give you the same motion quality without the refusals.

Qwen 3.5 is Alibaba's open-weights LLM, and its 27B "Writer" fine-tunes are the strongest uncensored models for long-form scenes at their size. We compare them with Gemma 4 roleplay tunes, explain what a writer tune optimises for, and show where each belongs in a chat product.

Krea's real-time generation and enhancement stack is the fastest way to iterate on a character image, and its vendor filter refuses exactly the images adult creators want to iterate on. We explain where the filter sits, what changes when the same pipeline runs uncensored, and how it compares with a Flux workflow.

Voice is the part of adult AI that hosted vendors police most strictly, which makes open-weights TTS the only workable route. Higgs Audio is the model we run: zero-shot cloning from a short clip, expressive delivery, and no filter on what it says. Here is how it sounds, how it clones, and where it still falls short of ElevenLabs.

ElevenLabs makes the most natural synthetic voices available and prohibits sexually explicit content in its terms. We walk through where the line sits in practice, what an adult platform can still use it for, how enforcement is set up, and why the answer for explicit voice is an open-weights model beside it.

Image-to-video is the adult creator's workflow, and the reference frame decides more of the result than the prompt does. This guide covers how upload classifiers score your image, why they reject clothed pictures, and the preparation steps that make Wan 2.7 and Seedance 2.0 animate a frame cleanly.