The Rise of AI Influencers
and how to create your own AI characters
This little 5 minute video is the Cliff’s Notes on how to build NSFW AI influencers.
Aitana, the subject of the video, is ostensibly a 25-year-old model from Barcelona who likes to play video games and pose for provocative selfies, but her name is a cheeky nod to her real identity. In fact, her creators don’t even try to hide her artificial-ness; it’s right there in her Instagram bio.
That honesty hasn’t stopped Aitana from bringing in the cash.
According to EuroNews, Aitana earned her creators €3,000-10,000 a month as of last November, when she had only 121,000 Instagram followers. Aitana currently has 307,000 followers on Instagram and an unknown number of paid subscribers on Fanvue (an AI-friendly OnlyFans clone), so her revenues are surely higher now.
Not exactly “getting filthy rich” as the title suggests, but definitely a good income.
There are other examples, too, some of which are less forthcoming about their AI identities.
Our take: the paperclipping of the OnlyFans economy
What does this portend for actual human influencers?
The biggest impacts will probably be felt in the adult industry. There’s much more variety behind the appeal of influencers in other niches, which will likely protect those influencers from AI competition longer, if not necessarily indefinitely.
How will actual humans compete when AI creators can make hundreds of adult photos or videos in dozens of different locations and costumes in a few hours?
It seems likely that the marginal cost of adult content will trend towards zero.
The only two really obvious value-adds for real girls are 1) personalized messages, chats, videos, requests, etc., and 2) spending time together in person (e.g. prostitution, or maybe fan meetups).
Number 1 can already be done with AI, if somewhat clumsily. In fact, successful creators are already using AI to chat with their fans. That leaves prostitution (which is already the main source of income for a lot of OnlyFans creators) and fan meetups as differentiators.
A lot of people could be affected.
OnlyFans doesn’t release data about its content creators, so it’s difficult to find reliable statistics on the site. Some market research sites say OnlyFans had around 2 million content creators in 2021 and 3.18 million in 2022 – a surge likely driven by Covid lockdown rules and pandemic stimulus spending. It’s safe to assume that the vast majority of these creators are making adult content. Which means that potentially millions of people – mostly young women – could lose their side hustle income or primary income in the next few years.
Incidentally, this also means that the time to profit from AI influencing is now. The window to make money here – on a small scale, at least – is going to close rapidly. Expect well-funded tech companies and studios to dominate AI influencers on existing mainstream platforms. As for adult content, the next OnlyFans will likely cut out not just real female creators, but also content creators of all kinds. It probably won’t be a platform at all; it’ll more likely be a 100% in-house produced, personalized, AI-driven experience.
How to make your own AI influencer
In the meantime, there is opportunity.
There are online tools based on Stable Diffusion that could be used to create an AI influencer, but they all have limitations either in terms of training, prompt sophistication, cost, non-NSFW guardrails, or a combination of those factors.
Midjourney
The most well-known is Midjourney, which is a dedicated image generator. Midjourney plans start at $10/month. The results generally seem pretty good, but there are some strong guardrails.
RenderNet
RenderNet is specifically designed to make AI characters. The killer feature here is that the app keeps the face consistent in every image.
This video shows how to use RenderNet for still images:
pincel
Another dedicated AI character generator is pincel. pincel plans start at $19/month. There’s a lot less info out there on pincel, but it’s still worth a look.
Stable Diffusion on a Local Machine
Another approach is to use Stable Diffusion locally or in a cloud environment, like Google Colab. Automatic1111 seems to have the most buy-in at the moment in terms of community and accumulated knowledge. The learning curve is steep, but the cost is free (if your machine meets the system requirements), prompts can be extremely sophisticated, training can get very specific, and there are no guardrails.
Check out these guides to install and set up Automatic1111 on Windows or Mac. Forge and ComfyUI are also gaining a lot of traction.
For a more thorough take on installation, check out this video:
Once set up, the real work begins. This guide covers LORA training – which is just one of the many things you’ll need to learn – on Automatic111:
Up next…
I plan to test each of these tools and methods over the next few weeks. Expect some deep dives.