[Journey] Building a €500/Month Income with an AI Girl

[LucasOFF]

You have to really fine-tune your prompts and test different models to see what works best. I recommend using Automatic1111 for testing, it’s faster and has a more visual interface. It also has the CivitAI browser built-in, so you can download models directly if you set up an API key in your CivitAI account

To create a LoRA, you’ll need a dataset with at least 100 images, depending on how consistent you want the output to be. I created separate datasets for the face and the body, then trained them separately. Once you have your safetensors file, you can add it to your generation workflow and call it by name to apply it to your images.

Absolute legend! Thank you, I appreciate it! I will look into Automatic1111, however I have some spare £ available so it only takes about 3 seconds to generate an image for me. Just learned about custom LoRAs on CivitAI, and it makes a huge difference. Just tweaking prompt and weights for loras have made a result that fooled some people I know.

To create body pics - do you try to generate as many types of the same body as possible, then train lora, and then call it when generating body pics? Do you have to faceswap at first to make it a bit consistent?

 

[CoveredEmbers]

Interesting, can you share more?

Sure, just ask me what you want to know and I’ll do my best to answer.

Absolute legend! Thank you, I appreciate it! I will look into Automatic1111, however I have some spare £ available so it only takes about 3 seconds to generate an image for me. Just learned about custom LoRAs on CivitAI, and it makes a huge difference. Just tweaking prompt and weights for loras have made a result that fooled some people I know.

To create body pics - do you try to generate as many types of the same body as possible, then train lora, and then call it when generating body pics? Do you have to faceswap at first to make it a bit consistent?

Yes, I generate as many similar body shots as possible and then carefully select the best images for the dataset. If some images don’t look right, I leave them out because they can ruin the LoRA’s quality. I’ve never done faceswap, the hardest part is building the dataset from scratch. Faceswap would definitely be faster, though.

 

[LucasOFF]

Sure, just ask me what you want to know and I’ll do my best to answer.



Yes, I generate as many similar body shots as possible and then carefully select the best images for the dataset. If some images don’t look right, I leave them out because they can ruin the LoRA’s quality. I’ve never done faceswap, the hardest part is building the dataset from scratch. Faceswap would definitely be faster, though.

Oh mate, you're a legend, wish I could donate some £ for a coffee for your efforts! And the last one before I go dive deep into this myself - do those body shots have faces or just body shots? If they have faces - does it mean you have to try collect the body dataset using the first LoRA with face to make sure the face stays consistent on non-portrait pics?

 

[andyat11]

Interesting, can you share more?

It gets quite complicated, my suggestion use a GPT to help you through the process.

 

[andyat11]

Thanks mate. I used to post about twice a day, but now it’s closer to once every two days. I’ve only made a few reels so far, mostly because I’ve been lazy about it. I should definitely do more because they reach more people. So right now, it’s mainly photos and stories.



Yes, it’s stated in my profile that it’s AI, I’ve never hidden that. Like I mentioned a few times already, it’s pretty obvious to anyone watching. I mainly convert with posts or stories.



I don’t have a specific guide to recommend, I watched a mix of tutorials and guides and did a lot of testing. I spent about $16 on RunPod (about 60 hours of VM time) just for experiments and training. I don’t really try to make them look real because my target is more about the character style. If I start working on the new model, I might have other tips to share later on, but for now, that’s all I’ve got.



I don’t really talk to them much on Instagram, I usually say it’s not the place to chat. The real work happens on Fanvue. From what I’ve seen, it’s better to have fewer but high-quality clients than a bunch who pay very little. Focus on quality rather than quantity. Of course, to scale faster you’ll need more traffic, but it has to be good traffic.

When you say watch your AI are you doing mostly videos or pictures.

 

[Genius Ranker]

Really solid start—love how detailed and honest your approach is. You're doing all the right things: testing, refining, and staying consistent. That custom LoRA setup is next-level.

 

[CoveredEmbers]

Oh mate, you're a legend, wish I could donate some £ for a coffee for your efforts! And the last one before I go dive deep into this myself - do those body shots have faces or just body shots? If they have faces - does it mean you have to try collect the body dataset using the first LoRA with face to make sure the face stays consistent on non-portrait pics?

Haha, no worries mate, we’re all here to help each other out. I did it in a way that I wasn’t super happy with, and if I had to do it over, I’d probably do it exactly how you said. First, train a close-up face LoRA (make sure to use different expressions and maybe some CivitAI LoRAs to help), and then build a body dataset. After that, use ADetailer (a setting in Automatic1111) to inpaint the face with your custom face LoRA for better consistency.

When you say watch your AI are you doing mostly videos or pictures.

If I understood the question right, I meant that my character is pretty obviously AI, so it’s clear from the start.

Really solid start—love how detailed and honest your approach is. You're doing all the right things: testing, refining, and staying consistent. That custom LoRA setup is next-level.

Thanks so much mate, I really appreciate the support!

 

[ensaimado]

Haha, no worries mate, we’re all here to help each other out. I did it in a way that I wasn’t super happy with, and if I had to do it over, I’d probably do it exactly how you said. First, train a close-up face LoRA (make sure to use different expressions and maybe some CivitAI LoRAs to help), and then build a body dataset. After that, use ADetailer (a setting in Automatic1111) to inpaint the face with your custom face LoRA for better consistency.




If I understood the question right, I meant that my character is pretty obviously AI, so it’s clear from the start.




Thanks so much mate, I really appreciate the support!

What type of post do you make? I mean, i'm generating amazing photos but I'm not being capable of doing a nice video to make a reel

 

[Plastic Bowl]

Looks like you're doing great man, keep up that progress

 

[GIDEON6155]

Good lck

 

[CitrusTit]

Great job man, happy for you!!

 

[CoveredEmbers]

Update:

Things have slowed down a bit these last few days. Right now I’m mainly trying to convert the free followers I’ve got on Fanvue into paying customers. If anyone has tips or tricks for that, I’d love to hear them.
I also found a TTS model I like for generating audio tracks, so I’m going to start making some audio content soon and see if that helps bump up the earnings.

Earnings so far: $3,338.88

What type of post do you make? I mean, i'm generating amazing photos but I'm not being capable of doing a nice video to make a reel

Reels don’t have to be actual videos, you can make them from images too. Just think outside the box, mate.

Looks like you're doing great man, keep up that progress

Hey Plastic, you’re the person who inspired me to start, so your comment really means a lot. Thanks so much mate. Do you have any tips on how to convert free followers on Fanvue into paying ones?

Good lck

Great job man, happy for you!!

Thank you guys, really appreciate it.

 

[Blade4Pro]

Hey, awesome journey so far! I've been hesitant to start the same journey because i thought AI OFM or fvm lol is oversaturated, but i'll give it a go. When you created your instagram account, did you go through the process of warming it up, or did you make an account and just started posting right away? Also what do you think about these for traffic generation: Instagram, Threads, Tiktok and Reddit, and do you only have one ai creator so far, or are you working with multiple accounts?

 

[LucasOFF]

Update:

Things have slowed down a bit these last few days. Right now I’m mainly trying to convert the free followers I’ve got on Fanvue into paying customers. If anyone has tips or tricks for that, I’d love to hear them.
I also found a TTS model I like for generating audio tracks, so I’m going to start making some audio content soon and see if that helps bump up the earnings.

Earnings so far: $3,338.88



Reels don’t have to be actual videos, you can make them from images too. Just think outside the box, mate.



Hey Plastic, you’re the person who inspired me to start, so your comment really means a lot. Thanks so much mate. Do you have any tips on how to convert free followers on Fanvue into paying ones?




Thank you guys, really appreciate it.

Oh mate that's awesome! You've already on over €500 per month on average haha, so could be considered this goal reached! Thinking of a new goal maybe?

I've spent all weekend figuring out how this all works and managed to get a very realistic face with some LoRAs from civitai but I can't seem to be able to make her smile or move her head to other direction etc because I want to collect a face pack of 100+ pics and train custom LoRA on her. Now looking up how I can fix the issue of no emotion/no head rotation lol The model I'm using is SDXL the basic one, but it starts to feel to be quite a limited one? Would you mind sharing what model are you using? Have you tried Flux or other models?

 

[CoveredEmbers]

Hey, awesome journey so far! I've been hesitant to start the same journey because i thought AI OFM or fvm lol is oversaturated, but i'll give it a go. When you created your instagram account, did you go through the process of warming it up, or did you make an account and just started posting right away? Also what do you think about these for traffic generation: Instagram, Threads, Tiktok and Reddit, and do you only have one ai creator so far, or are you working with multiple accounts?

Actually, I didn’t have any experience before this journey, I’d only made about $2.5k four years ago with CPA games, so I was basically starting from zero. I started by making one account and posting lightly at first, following just a few people manually each day. Later on, I made a “slave” account to funnel traffic to the main one.
I think Reddit can be good if your model is super realistic. So far I’m only working with one AI creator.

Oh mate that's awesome! You've already on over €500 per month on average haha, so could be considered this goal reached! Thinking of a new goal maybe?

I've spent all weekend figuring out how this all works and managed to get a very realistic face with some LoRAs from civitai but I can't seem to be able to make her smile or move her head to other direction etc because I want to collect a face pack of 100+ pics and train custom LoRA on her. Now looking up how I can fix the issue of no emotion/no head rotation lol The model I'm using is SDXL the basic one, but it starts to feel to be quite a limited one? Would you mind sharing what model are you using? Have you tried Flux or other models?

Probably I’ll start another journey, but right now my focus is making the income stable and automating the process as much as possible, it still takes a lot of time and work for now.
I’m using SD1.5 (Stable Diffusion). It’s older, but I think there are way more LoRAs and embeddings for this version. SDXL didn’t really work for me, and Flux feels too new. Be careful not to mix them, if you mix SDXL and Flux or SD, the results usually aren’t great. If you’re aiming for hyper-realistic results, I’d recommend looking at the CyberRealistic SD1.5 model.

 

[ZenovaStrategies]

Good Luck on your journey. Hope this doesn't get ALOT of competition like in other fields.

 

[LucasOFF]

Probably I’ll start another journey, but right now my focus is making the income stable and automating the process as much as possible, it still takes a lot of time and work for now.
I’m using SD1.5 (Stable Diffusion). It’s older, but I think there are way more LoRAs and embeddings for this version. SDXL didn’t really work for me, and Flux feels too new. Be careful not to mix them, if you mix SDXL and Flux or SD, the results usually aren’t great. If you’re aiming for hyper-realistic results, I’d recommend looking at the CyberRealistic SD1.5 model.

That's a nice shout, thanks for the suggestion! This is exactly my experience - it's hard to make it work and not enough LoRAs there for realism, the existing ones make the model nearly impossible to modify. I will try to move to SD 1.5 and check it out. Might have to look for a new keeper face even though the one I just generated is so good I'm personally drooling over her lol

Regarding automation, I would suggest looking into Python. As a SWE myself in Python I can say it's the easiest language you can read as well as LLMs are good for micro dumb tasks with it, so you can automate some things yourself. I'm looking into it as well from the start and I see that both Automatic1111 and ComfyUI have APIs available.

 

[Blade4Pro]

Actually, I didn’t have any experience before this journey, I’d only made about $2.5k four years ago with CPA games, so I was basically starting from zero. I started by making one account and posting lightly at first, following just a few people manually each day. Later on, I made a “slave” account to funnel traffic to the main one.
I think Reddit can be good if your model is super realistic. So far I’m only working with one AI creator.



Probably I’ll start another journey, but right now my focus is making the income stable and automating the process as much as possible, it still takes a lot of time and work for now.
I’m using SD1.5 (Stable Diffusion). It’s older, but I think there are way more LoRAs and embeddings for this version. SDXL didn’t really work for me, and Flux feels too new. Be careful not to mix them, if you mix SDXL and Flux or SD, the results usually aren’t great. If you’re aiming for hyper-realistic results, I’d recommend looking at the CyberRealistic SD1.5 model.

Do you just repost content from your main account to the slave accounts, or is it new content for each, each time you post? Also i found some tool from one guy's tutorial that charges 0.30$ per hour for using but i didn't try it yet, i dont know if i should try using his *tool* for generating AI content or if i should follow your principles, so far i understood nothing from what you've been saying about generating content yourself, i'll read again until i understand lol.

 

[Plastic Bowl]

Update:

Things have slowed down a bit these last few days. Right now I’m mainly trying to convert the free followers I’ve got on Fanvue into paying customers. If anyone has tips or tricks for that, I’d love to hear them.
I also found a TTS model I like for generating audio tracks, so I’m going to start making some audio content soon and see if that helps bump up the earnings.

Earnings so far: $3,338.88



Reels don’t have to be actual videos, you can make them from images too. Just think outside the box, mate.



Hey Plastic, you’re the person who inspired me to start, so your comment really means a lot. Thanks so much mate. Do you have any tips on how to convert free followers on Fanvue into paying ones?




Thank you guys, really appreciate it.

Glad it inspired you man! I think the best way are lots of promotions and give them some teasing content (blur or lingerie) every once in a while. It's also a good way to boost likes on your account.

 

[BeyondProxies]

Hello, I bet you could improve your success by trying out different marketing strategies and lowering the price