[Journey] From $0 to $10k a month with AI OFM [with Fanvue]

Akaroth

Akaroth

Senior Member
Jr. VIP
Joined
Mar 1, 2021
Messages
1,111
Reaction score
297
Hi BHW.

This is the beginning of my journey from $0 to $10k profit per month with AI OFM.

I tried to start this journey 1 year ago, but I ran into some problems with my other businesses and had to put it off.

But now I'm back. I bought a RTX 5090, and about 5 weeks ago, I started learning and experimenting with a lot of things in ComfyUI to generate AI content.

What do I want to do in this journey?

I want to create an influencer, create social media accounts, drive traffic to Fanvue, and make $10k profit per month.

To be honest, I started this project 5 weeks ago, but when I started reading @shijin journey, I thought it would be a good idea to do one myself.

So I’ll be sharing progress, numbers, mistakes, and lessons along the way. I'll post at least once a week.

Where am I right now?

In these 5 weeks, I created the face of the first model I'll manage and built a dataset to train a LoRA to get consistency in her face.

I’m still having some problems with the training. The consistency isn’t 100% there yet, but I feel like I’m getting close.

So what's next?

Right now, my main focus is:

1. Get consistency in SFW & NSFW content.
2. Start posting content on social media.

And I have to do a ton of other things, but those two alone will probably keep me busy for a while.

When will I hit $10k/month?

Honestly, I don’t know.

I’m dedicating around 6–7 hours per day, Monday to Saturday, to this project. I’d love to be making $10k profit per month by the end of this year, but I know there’s a lot to learn, test, and improve before that happens.

Wish me luck guys :D
 
Advertise on BHW
Kindly share the set up and process I can join in and create my model too since I'm interested
 
solid start, but the lora consistency is the easy 20%. the $10k doesnt live in render quality, it lives in the close. a perfect looking model still earns ~nothing if the click lands on a passive subscribe page instead of a conversation. fanvue traffic that just "subscribes" and gets a wall of posts churns fast, the money is in the dm back-and-forth that gets them to a paid unlock. whats your plan to turn a fanvue click into an actual chat, are you running the dms yourself or have you thought about how that scales past the first few fans?
 
marcotwzrd2 said:
solid start, but the lora consistency is the easy 20%. the $10k doesnt live in render quality, it lives in the close. a perfect looking model still earns ~nothing if the click lands on a passive subscribe page instead of a conversation. fanvue traffic that just "subscribes" and gets a wall of posts churns fast, the money is in the dm back-and-forth that gets them to a paid unlock. whats your plan to turn a fanvue click into an actual chat, are you running the dms yourself or have you thought about how that scales past the first few fans?
Click to expand...
Thank you for your comment!

I did a big research 1 year ago, and I know that most of the money comes from chatting. I think I'll handle the chats by myself with a huge help of the AI, as @shijin did, and I'm planning to create an automated bot that handles the chats too, but honestly, I haven’t thought about it much because I’m still in the very early stages of this project.

Maybe creating a good AI chatter is too hard, and I'll end up hiring someone to handle the chat. I don't know; there's still a long way to go.
 
yeah the chatter is where it actually gets decided, and the 'too hard' instinct is right but for the wrong reason. a good ai chatter isnt hard because of the model, gpt-class stuff holds a convincing convo fine now. its hard because of the operational layer around it: knowing when to stop chatting and drop the paid unlock, handling the guy who claims he already paid, not breaking character when someone tests it, and routing the payment without a human touching it. that orchestration is 90% of the work and 0% of the comfyui fun. one number worth planning around: most buyers who convert do it inside the first ~8 minutes of the first chat, so the bots job isnt to be a soulmate, its to read intent fast and make the ask before the novelty fades. if you hire it out youll hand 10-20% rev share to an agency chat team and lose the margin you bought the 5090 to keep. building it yourself is very doable, just budget way more time for the close-logic than the persona. whats the payment side look like, fanvue native checkout or are you routing to something you control?
 
marcotwzrd2 said:
yeah the chatter is where it actually gets decided, and the 'too hard' instinct is right but for the wrong reason. a good ai chatter isnt hard because of the model, gpt-class stuff holds a convincing convo fine now. its hard because of the operational layer around it: knowing when to stop chatting and drop the paid unlock, handling the guy who claims he already paid, not breaking character when someone tests it, and routing the payment without a human touching it. that orchestration is 90% of the work and 0% of the comfyui fun. one number worth planning around: most buyers who convert do it inside the first ~8 minutes of the first chat, so the bots job isnt to be a soulmate, its to read intent fast and make the ask before the novelty fades. if you hire it out youll hand 10-20% rev share to an agency chat team and lose the margin you bought the 5090 to keep. building it yourself is very doable, just budget way more time for the close-logic than the persona. whats the payment side look like, fanvue native checkout or are you routing to something you control?
Click to expand...
Thank you for your comment!

I spent around 3 hours yesterday researching about the chatting, and yes, I found out that the orchestration is most of the work (I didn't know that hehe)

So, I think I'll do something like this:
1. Receive a message
2. Save it to the database
3. Create/update the fan's profile
4. Read the history (last 10 messages or something like that)
5. Classify the intent
6. Check what's the best tone or mood to use with this fan
6. Generate a response using AI (Finally, here's when the trained AI appears :P )
7. Evaluate the message (using AI too)
8. Send it for review or auto-send

And yes, I was thinking in adding something that triggers a message to me when there's a signal alarm, something like someone asking if it's an AI, but I didn't think about those payment problems. Thank you for letting me know.

Anyway, I thought it was going to be easier tbh, so I'm going to start with a version that's quick and easy to make. I won't spend time on this yet.

I'll probably start building the full version when I get some traffic. Until then, I'll handle it by myself (with help of AI)
 
why do you need a RTX 5090? isnt it expensive, what software/tutorials have helped you making models?
 
that pipeline is basically right for 3 hours of research, nice. one trap in the "quick version now, real one when i get traffic" plan: the per-fan state (your steps 3 and 5) is the part you cant cheaply bolt on later. a v1 that just generates replies with no memory of where each fan sits gets rewritten, not extended, because the persona and the close end up coupled once you add them. even a dumb state from day one, has this fan been asked, did they pay, last unlock, makes the upgrade painless instead of a teardown. and keep auto-send off until the intent classifier stops mislabeling the "is this a bot" testers, those are exactly the fans that burn when it answers wrong. are you persisting per-fan state in v1 or starting stateless?
 
meowboy09 said:
why do you need a RTX 5090? isnt it expensive, what software/tutorials have helped you making models?
Click to expand...
You can do it by renting servers instead of buying an RTX 5090.

One year ago, I saw a course on how to use ComfyUI. Around 40 hours each. To be honest, that was a waste of time. If you wanna learn how to use ComfyUI, there's a 4hs video of ComfyUI made by Pixaroma on Youtube, and then start researching on Reddit/CivitAI some workflows. And then, you have to test and test and test until the output is good.

marcotwzrd2 said:
that pipeline is basically right for 3 hours of research, nice. one trap in the "quick version now, real one when i get traffic" plan: the per-fan state (your steps 3 and 5) is the part you cant cheaply bolt on later. a v1 that just generates replies with no memory of where each fan sits gets rewritten, not extended, because the persona and the close end up coupled once you add them. even a dumb state from day one, has this fan been asked, did they pay, last unlock, makes the upgrade painless instead of a teardown. and keep auto-send off until the intent classifier stops mislabeling the "is this a bot" testers, those are exactly the fans that burn when it answers wrong. are you persisting per-fan state in v1 or starting stateless?
Click to expand...
Yes, I’m going to save the fan state since the 1st day, so then I can train the AI with it. I think I’ll store the basics: current stage, last intent, last offer sent, whether they purchased, message count, and risk flags.

That's good advice, thank you! :D
 
Akaroth said:
You can do it by renting servers instead of buying an RTX 5090.

One year ago, I saw a course on how to use ComfyUI. Around 40 hours each. To be honest, that was a waste of time. If you wanna learn how to use ComfyUI, there's a 4hs video of ComfyUI made by Pixaroma on Youtube, and then start researching on Reddit/CivitAI some workflows. And then, you have to test and test and test until the output is good.


Yes, I’m going to save the fan state since the 1st day, so then I can train the AI with it. I think I’ll store the basics: current stage, last intent, last offer sent, whether they purchased, message count, and risk flags.

That's good advice, thank you! :D
Click to expand...
So for people who don't have a spare 10k spare for a RTX 5090 you can just rent a VPN to run comfyui on? do you have any recommendations
 
thats a clean schema, youve got the operational layer covered. one field that pays for itself: total_spent + last_purchase, basically a running ltv per fan. the biggest lever isnt converting new fans, its the bot spotting a whale and escalating the ask instead of charging everyone the same unlock, a guy whos dropped 200 should never see the first-timer price. and a last_active stamp so you can fire a win-back before they churn silently, thats where most of the lost sub revenue actually goes. when the dms scale past what you can read by hand thats the real wall, happy to compare notes when you get there. whats the first traffic source going to be, ig reels?
 
Akaroth said:
Hi BHW.

This is the beginning of my journey from $0 to $10k profit per month with AI OFM.

I tried to start this journey 1 year ago, but I ran into some problems with my other businesses and had to put it off.

But now I'm back. I bought a RTX 5090, and about 5 weeks ago, I started learning and experimenting with a lot of things in ComfyUI to generate AI content.

What do I want to do in this journey?

I want to create an influencer, create social media accounts, drive traffic to Fanvue, and make $10k profit per month.

To be honest, I started this project 5 weeks ago, but when I started reading @shijin journey, I thought it would be a good idea to do one myself.

So I’ll be sharing progress, numbers, mistakes, and lessons along the way. I'll post at least once a week.

Where am I right now?

In these 5 weeks, I created the face of the first model I'll manage and built a dataset to train a LoRA to get consistency in her face.

I’m still having some problems with the training. The consistency isn’t 100% there yet, but I feel like I’m getting close.

So what's next?

Right now, my main focus is:

1. Get consistency in SFW & NSFW content.
2. Start posting content on social media.

And I have to do a ton of other things, but those two alone will probably keep me busy for a while.

When will I hit $10k/month?

Honestly, I don’t know.

I’m dedicating around 6–7 hours per day, Monday to Saturday, to this project. I’d love to be making $10k profit per month by the end of this year, but I know there’s a lot to learn, test, and improve before that happens.

Wish me luck guys :D
Click to expand...
You can simply rent GPU from runpod etc and the hard part is to train properly the model to get good results, which takes time. How you do with reddit? Do you have your own groups?
 
meowboy09 said:
So for people who don't have a spare 10k spare for a RTX 5090 you can just rent a VPN to run comfyui on? do you have any recommendations
Click to expand...
Most people use RunPod. To be honest, I didn't use it enough, and maybe there's something better. That's the only one I know.

marcotwzrd2 said:
thats a clean schema, youve got the operational layer covered. one field that pays for itself: total_spent + last_purchase, basically a running ltv per fan. the biggest lever isnt converting new fans, its the bot spotting a whale and escalating the ask instead of charging everyone the same unlock, a guy whos dropped 200 should never see the first-timer price. and a last_active stamp so you can fire a win-back before they churn silently, thats where most of the lost sub revenue actually goes. when the dms scale past what you can read by hand thats the real wall, happy to compare notes when you get there. whats the first traffic source going to be, ig reels?
Click to expand...
That's another good advice, thank you, I'll write it down right now.

I want to start posting on IG (reels), TikTok, and Reddit simultaneously. I'll see how they work.

AnnSo said:
You can simply rent GPU from runpod etc and the hard part is to train properly the model to get good results, which takes time. How you do with reddit? Do you have your own groups?
Click to expand...
I'm not doing anything with Reddit yet. I just read posts on subreddits to learn some things about the AI and be updated.
 

UPDATE #1​


I'm finally getting consistency in the influencer's face. Sometimes I have problems with some little moles on her face, and I've been trying to fix this for the last 2 days, but it isn't a big deal. The important thing is that her face's structure is always the same. I'd love to fix that (and I'm trying to do that right now), but if it's too hard, I could just remove the moles and get into the next steps.

What are my short-term goals?

• Train her body (SFW and NSFW) for consistency on images.
• Train a LoRA for consistency on videos.
• Create videos with a high level of realism.

Once I've accomplished that, I'll finally be able to create content and start thinking about the traffic.
 
the mole thing is the perfectionism trap, fans never zoom in hunting for a mole, they only notice when she looks like a different person between posts. youve already got the face structure locked so honestly ship it, traffic will tell you fast what actually matters and its never the mole. one thing on launching ig + tiktok + reddit at the same time: with a brand new model thats three different warm-up and ban profiles to debug at once, youll spend more time on account health than on content. id get one stable first then clone the playbook to the next. which one are you leaning toward as the main?
 
marcotwzrd2 said:
the mole thing is the perfectionism trap, fans never zoom in hunting for a mole, they only notice when she looks like a different person between posts. youve already got the face structure locked so honestly ship it, traffic will tell you fast what actually matters and its never the mole. one thing on launching ig + tiktok + reddit at the same time: with a brand new model thats three different warm-up and ban profiles to debug at once, youll spend more time on account health than on content. id get one stable first then clone the playbook to the next. which one are you leaning toward as the main?
Click to expand...
Yes, I know that it's not that important, but I wanted to see if I understand what's the setting that's messing with it, or if it's something wrong in the dataset I'm using.

I would like to have it because it adds a little bit more of realism, but yes, I'll move on.

I'll go with those 3 simultaneously (at least in the beginning) because IG Reels and TikTok use almost the exact same content, and I already created the accounts 2 weeks ago, so I think both of them should be good for posting right now. And about Reddit, I've been selling Reddit upvotes and comments since 2023, so it's like the social media that I already know.

Anyway, I don't have experience with Instagram and TikTok, so I'll go slowly to avoid bans.
 
honestly your reddit background is the biggest asset in that list and youre treating it like the side one. for this funnel reddit beats ig/tiktok on the part that matters: it allows the actual nsfw content natively and the traffic is already in buying mode, youre not fighting an algo to stay sfw then praying they tap a bio link. ig/tiktok are top-of-funnel reach you have to launder into a click, on reddit you post the content and drop the link in the same place. and you already run aged accounts + upvotes, so you can seed the early velocity most people spend months earning. id make reddit the main and let ig/tiktok feed it. are your existing accounts aged enough for the bigger nsfw subs or would you warm fresh ones for this?
 
Advertise on BHW
You must log in or register to reply here.
Sign up now!

Main Menu

Home Main Forum List Black Hat SEO White Hat SEO BHW Newbie Guide Blogging Black Hat Tools Social Media Downloads

Marketplace

Account Selling Content / Copywriting EBooks / Methods Hosting Misc Proxies For Sale SEO - Link building SEO - Packages Social Media Social Media - Panels Web Design More

Making Money

Affiliate Programs Hire a Freelancer IM Journeys Making Money Pay Per Click Site Flipping Want to Buy

BlackHat World

Dispute Resolution Forum Suggestions & Feedback Introductions Lounge
Back
Top