4chan Users Are Sharing Tools and Tips for Homebrewing AI Deepfake Porn

Users on 4chan’s “Technology” board have shown heightened interest in tools and techniques for using generative artificial intelligence models to create their own “deepfake porn” in recent months, new analysis from Open Measures and the American Sunlight Project has found.

The uptick followed the abrupt closure of 4chan’s “Adult Requests” board in May, which functioned as one of the internet’s most active outposts for synthetic non-consensual intimate imagery (NCII) abuse prior to its removal, and the implementation of the TAKE IT DOWN Act in the United States, which criminalized the online distribution of NCII and imposed penalties against platforms that failed to respond to removal requests.

Amid these shifts, Technology board users have circulated dozens of different “guides” containing information about AI models and using them to generate synthetic images. Most guides consisted of compiled hyperlinks to commercial and open-source AI tools, instructions for using them offline on personal devices, and repositories of code for low-rank adaptation models (“LoRAs”) that tweak existing AI models for specific use cases.

While those guide posts rarely made explicit mentions of using the provided resources to produce synthetic NCII, we found that threads where the guides were shared regularly elicited replies from 4chan users who indicated they were interested in doing exactly that.

Increased Chatter About Generative AI Tools on 4chan’s ‘Technology’ Board Propelled, In Part, By Users Interested in Making Synthetic NCII

The number of posts per day on 4chan’s Technology board discussing generative AI models and synthetic imagemaking increased after the closure of Adult Requests and the implementation of the TAKE IT DOWN Act in May. Our analysis found that the increase is attributable, in part, to users who indicated they were interested in using AI models to create sexually explicit synthetic images.

A search query containing 41 terms related to AI image creation, including the names of specific AI models and techniques, identified more than 36,000 posts on the Technology board between May 20 and Sept. 8, 2026. Using the Timeline tool in Open Measures’ platform to visualize the number of posts per day that included at least one term from our query, we saw that discussion of those models and methods increased in late June and spiked in early August.

Caption: A Timeline chart shows the number of posts per day on 4chan’s Technology board that mentioned at least one of 41 terms related to AI imagemaking between May 20 and Sept. 8, 2026.

Talk of generative AI models on the Technology board spanned many different use cases during the period we analyzed, ranging from benign to malicious. To identify posts that discussed AI image creation in the context of NCII, we created three expanded search queries to identify posts that also included terms related to sexually explicit imagery.

Viewing the results of those searches over the same timeframe, we saw that periods of increased chatter about generative AI models alongside terms related to explicit imagery often coincided with periods of increased interest in the tools, broadly.

Caption: A Timeline chart shows the number of posts per day on 4chan’s Technology board that mentioned at least one of 41 terms related to AI imagemaking and additional terms related to sexually explicit imagery between May 20 and Sept. 8, 2026.

We used Open Measures’ Activity tool to identify which URL domains frequently appeared in posts matching our search query. Hyperlinks included in posts discussing AI image generation on the Technology board most often directed to pages on the developer platform GitHub, text editing site Rentry, research database arXiv, machine learning development forum Hugging Face, and anonymous file hosting service Catbox. This suggests that conversation about generative AI tools on the board included discussions of technical details related to their use, and an informal pipeline of sorts.

For instance: arXiv hosts thousands of preprints and detailed papers about principles, methods, and vulnerabilities in machine learning. HuggingFace offers publicly accessible downloads of both base and trained models, image datasets, and LoRAs that already exist of politicians and celebrities. GitHub enables modifications, jailbreaks, and other creative potential that is used to tweak models. Rentry allows users to train prompts and model skills, with results shared on anonymous media hosting sites like Catbox.

Caption: An Activity chart shows which URL domains appeared most often in posts on 4chan’s Technology board that mentioned at least one of 41 terms related to AI imagemaking between May 20 and Sept. 8, 2026.

‘It’s Laughably Easy’: Users Share and Solicit Advice for Generating Synthetic NCII on Guide Threads

The threads we identified on the Technology board sharing guides for generating images with AI models rarely include direct references to synthetic NCII, but discussions on those threads frequently attracted users who expressed interest in using them for that purpose.

Our expanded search queries that included terms related to sexually explicit imagery identified posts where Technology board users shared and solicited advice for making pornographic content with generative AI models. For example, in an Aug. 28 thread, after one user suggested a particular AI model wasn’t ideal for creating sexually explicit images, another user suggested they utilize a LoRA model for that purpose:

it’s laughably easy to just apply a nsfw Lora and go from there

In similar “guide” threads, Technology board users praised specific AI models for their lack of guardrails preventing the generation of sexually explicit content, trading links to custom LoRAs tailored to depict realistic nudity, and sharing tips for prompting models to produce synthetic NCII. Some posts in those threads sought advice on how to circumvent guardrails built into certain AI tools, as seen in this Aug. 28 post:

What is the best AI system or model currently for undressing women and generating lingerie / nude shots? I haven’t been able to create anything since Grok was neutered

Other posts we identified showed users asking for help using specific models to produce pornographic images, like this reply to a guide thread on Aug. 9:

I need a guide for minimax. I don’t know how it is supposed to be good at porn. All my gens are absolute SHIT! Including those with video references

Anti-NCII Laws Haunt Users Discussing Synthetic NCII on 4chan

Adult Requests disappeared two days after the implementation of the TAKE IT DOWN Act, a US law that criminalized NCII and mandated tech companies to respond to requests to remove it when flagged on their platforms. Alongside broader discussions of AI image generation, we also saw that users on the Technology board acknowledge the increased legal implications surrounding deepfake porn.

Two days after Adult Requests’ removal, one user on the Technology board seemed to mock the implementation of the TAKE IT DOWN Act in a May 23 post referring to President Donald Trump and former Vice President Kamala Harris:

Do you think if *an individual* were to keep posting AI-generated deepfake nudes of Trump being pegged by Kamala…on /pol/ and reporting their own posts to the FBI, that board would get shut down just like they have shut down [Adult Requests] earlier today?

If the said individual were to also retweet such threads to President Donald Trump on TruthSocial, would that speed things up?

Posts remarking on the removal of Adult Requests also appeared on other boards across 4chan, where users were seen sharing theories for why site moderators took down Adult Requests.

On May 22, in reply to a thread asking where Adult Requests went, a 4chan user hypothesized that the board had been removed in response to the implementation of the TAKE IT DOWN Act and a news report about the synthetic NCII the board’s users were producing and sharing:

some suspect it might be connected to this story.

https://archive.is/2026.05.21-102104/https://www.wired.com/story/4chans-misogynist-wizards-are-nudifying-women-by-request/

also the Take it down act deadline expired on may 19th:

https://en.wikipedia.org/wiki/Take_It_Down_Act

Throughout June and July, remarks like these, and direct user-to-user interventions within threads containing synthetic NCII, occurred sporadically on the Technology board.

Fast forwarding to early August, we observed a notable interaction in a continuation of a long-running megathread called “Local Diffusion General.” On Aug. 4, users discussed the release of an uncensored version of H3 Minimax — a powerful generative AI model with minimal guardrails. In response to a post linking to a pornographic adaptation of the model, one user condemned others for publicizing the model’s malicious uses and insinuated that violative content was still being produced offline:

can you 4chan hackers stop ruining it for everyone? you’re going to get this shit banned if you keep genning problematic stuff and posting it online

Despite this attempt to shift the public discussion, other users in the thread replied in support of continuing to generate deepfake NCII. One declared:

Let them archive the models while we do what we must. THERE’S NO BREAKS ON THE GOONING TRAIN!

Another user blamed copyright-violating posts on Reddit for the increased enforcement against deepfake sexual abuse on 4chan:

From what I’ve seen reddit is a lot more likely to get it “”“Banned”“” with all the copyrighted stuff they post. Nobody gives a fuck about what people do on here.

Our Take

Communities that traffic in synthetic NCII have been forced to adapt their tactics in the wake of the implementation of the TAKE IT DOWN Act, increased guardrails on major commercial generative AI tools, and other factors that have made creating and sharing deepfake porn online more difficult and costly.

Our research suggests that one way communities have adapted is by developing and sharing an array of tools and techniques for homebrewing synthetic NCII with tweaked AI models run locally on personal devices. While this has manifested as a potential decrease in publicly available synthetic NCII, this does not appear to be the result of 4chan communities’ willful compliance with the law. Rather, it appears these communities are adopting new mechanisms to preserve their remaining online presence. This development presents new challenges for researchers tracking online abuse and those tasked with enforcing laws and platform guidelines prohibiting NCII.

While it remains to be seen exactly where former Adult Requests users will resettle online, our research found that 4chan users’ interest in finding new methods of producing synthetic NCII, often by hosting modified AI tools locally on their personal devices, has increased in recent months. And as governments around the world enact new laws to criminalize the distribution of NCII, those capable of homebrewing synthetic NCII with AI tools may draw increased interest from other nefarious online actors and communities.

Open Measures and the American Sunlight Project will continue monitoring toxic online communities to track how AI deepfake abuse evolves.