Show HN: I Built an AI Tattoo Generator Using Flux | Hacker News Reader

Show HN: I Built an AI Tattoo Generator Using Flux

tattoopro.ai

39 pointsRyanwalker6472 comments

I am heavily tattooed and I tried generating a few designs. I'm sorry to say that everything it suggested was awful, and if an artist ever showed me any of this in their flash collection I would block them on Instagram and potentially call the police.

I asked it for "one hand stabbing another with an ornate dagger, traditional style." It got the style right-ish but everything else was terribly wrong. Daggers with multiple blades, hands with hands on them, fingers with multiple fingernails, etc.

This could be useful if it can reliably get the core requirements correct though. I can see myself generating some ideas with this and taking them to an artist as a basis for them to start with.

OP: It would probably be good to have a "these aren't even close, please try again" button which allows for a couple free retries.

but here's the thing: why spend a bunch of time trying to formulate a prompt to give to a machine to generate a bad image to bring to an artist who would then refine it, when you could just say to the artist "one hand stabbing another with an ornate dagger, traditional style" and get what you want in one shot?

One-time payment is nice! The other AI Tattoo services I've seen all offer monthly subscriptions which just doesn't make sense.

Price is in the ballpark, but a bit steep for me when I can run models on my own hardware. However, for people that don't want to set up a local model for whatever reason, it seems reasonable.

How much do you let people fiddle with the parameters? It would be nice if, for example, when I find a good-ish image that I could lock-in the seed or CFG and slightly tweak the prompt.

I wish this is something Ai services would understand. You don't charge bases in your teach stack. You charge based on your use case.

Tattoos are not something that you get every day so it makes no sense to pretend that your customers will use it like chat gpt. Costing more makes sense since it's a one off transaction that needs to cover overheads.

There is a whole raft of services which would seriously benefit from doing the same.

Undermind is one I can't justify buying because I do two literature reviews a month at most and have no use for an ongoing subscription I'll forget about. But I'd be happy to pay $5 per search.

I’m curious why you don’t think tattoo shops would want this as a monthly expense? $30/mo is cheap to give customers the ability to easily generate designs. Like another commenter said in this thread, good artists are often booked months out and that’s just to get the design started. Make a simple app that shows the UI in kiosk mode on an iPad using the store’s account. Even offer a finetuning service that they can use with photos of their own designs to get a unique style per customer (I’ve successfully finetuned myself with under 20 photographs, although style might be better with 50-100).

The other big benefit of tattoo shops is that they’re a well established type of business that you can sell to, instead of trying to market to consumers. First thing I’d do is buy a list of tattoo parlors and their contact information from InfoGroup/Data Axle/Dun & Bradstreet and start cold contacting them with a free month so they can try it out. I’d monitor the demos and reach out to those who don’t use their free month much to get feedback and refresh the demo period so they can try again. There are conventions where you can sell to the owners and artists en masse too. Google says there are over 20,000 stores in the US which is enough scale to just have a few GPUs to handle base load and autoscale during peak to drive costs down.

I think it can be a very lucrative lifestyle business to someone with good distribution and sales sense, just as long as some VC doesn’t come in and prop up a better resourced competitor.

I like the idea of locking in the seed and params. Will see what I can do

> for people that don't want to set up a local model for whatever reason, it seems reasonable.

The reason being that in order to get a GPU that doesn't commit sudoku the moment you install SD, you need to shell out way more than the price of a single tattoo.

That's why people who own many 3d printers make money. If all it takes to print a model is one 3d printer, why doesn't literally every single person on the planet own one?

For sure, cost is maybe the most common reason.

But I have friends who game on cards that can run SD no problem (prior to a recent upgrade, I had SD running on a 2060S with no problem, which is like $250?), they just don't want to for other reasons (laziness, hard drive space, feel that it is too complicated, etc.).

> commit sudoku

I hate it when my GPU decides to waste time on puzzles instead of rendering things.

Honest question, as someone who has thought of making an "AI wrapper" app myself - why would I use this rather than go to Gemini/ChatGPT/StableDiffusion/etc and prompting it myself?

Yeah I have the same issue with these types of projects as well. Could be interesting to map it on a 3D body part or scan so you could see how it looks on your body or next to your other tattoos.

I like this, will look into adding these features as it sounds fun to build

I’d recommend looking at how Photoshop does it. Those asset marketplaces sell template images that contain a layer that maps the user’s image onto the template surface, like for t-shirts and other printing product mockups.

https://creativecloud.adobe.com/discover/article/mock-up-a-t...

Will have a look, thanks!

I'd say it's definitely a personal preference.

Using a wrapper gives you a few benefits. - It lets you shortcut the time to having a refined prompt that gives you a somewhat reliable output

- Flux (like some models) don't have readily available interfaces as the model is usually required to be self-hosted. For TattooPRO I'm use Together.ai as they host Flux and I can then use their API instead of hosting it myself. The outcome is that users can then get a nice user interface to generate Tattoos with Flux and have some additional features like history and favorites to keep track of their generations.

I've also tried to make the experience as mobile-friendly as possible.

Hope the answers your question

Its not that theres no benefit at all, it's more like does it give me enough upside compared to something that is easy and free, doesnt require me trusting an app Ive never heard of, taking out my credit card, worrying about getting ripped off, etc.

But I could well be wrong, I wish you success.

Hehe, the non answer you got is spot on what I expected.

Here's the deal: AI tattoo generator, what could possibly go wrong? Liability. That is also why people want to pay, even if the dev cannot or tries not to be held accountable and even if it is for some Electron frontend for a customized prompt. Paying gives them [the (potential) customer] the feeling they get a worthy result. A lot of services work like this, btw, and it helps if the service is actually not cheap. Because why spend very little money on a tattoo design. You're worth it, right?

My take is simple. If you want a tattoo and CBA to do your own research (via a search engine, a professional tattoo artist, some kind of curated database, or gasp CAD it yourself like you'd do your 3D print) then ML-based search could be a viable, modern alternative but I would not want to get burned by '6 fingers' in hindsight. AI output needs to be qualified by a qualified human being, and you [random person who wants tattoo] are probably not said qualified human being. But could it aid a qualified human being? Absolutely, just a smaller customer base. So if you want to go for volume, you pretend to serve a customer base you cannot reasonably serve well.

Eh, I think it's a neat idea. No one's forced to use or buy this - as is the case with any offered service. Also, the 'qualified human being' in the end is still the tattoo artist who's actually doing the tattoo in this use case. I assume most people who would use this won't just get a 1:1 copy tattoo of an AI generated result, the artist can still reiterate and use the designs as a draft or inspiration.

I think it can make sense when you have some secret-sauce mixed in for whatever the application is. A custom fine-tuned model, text embedding, LoRAs, etc. It's certainly less convincing to me when someone offers just a plain wrapper around free/cheap/easily-accessible models.

But I can see the appeal of making it a bit easier for non-technical people when you add in surrounding features (favorites, history, etc.).

At this point even the fine tuning isn’t a big differentiator. It costs a few bucks to make one in Replicate and you don’t even have to caption the photos because it can use another model to do that (I usually download and improve them for the second run). You just upload a zip file of images and give it a keyword.

There’s an art to fine tuning but plenty of laypeople have done it, it just takes time to experiment and some cash for the cloud providers.

I think your definition of laypeople and my definition of laypeople are different. If I talked to anyone not in my IT department about fine-tuning, their eyes would glaze over in 2 seconds.

These types of services are, in my opinion, targeted at the people who live their entire computer lives in Chrome & Excel. Not people who know what fine-tuning is or can recognize what "Replicate" is without Google.

I don’t mean it’s common knowledge among laypeople, just that someone determined enough to spend a weekend reading image gen documentation and the StableDiffusion subreddit can probably figure it out. It’s not like they need to take a months long bootcamp to learn to code first. Once they sign up for replicate (and I guess github for SSO first), all they have to do is find the page for the fine tuning and upload a zipfile of images.

Its not that it has no appeal, its that I expect it to be a tough sell to get people to actually take out their credit card for this when its free and good enough to go to chat.com or gemini. But I may be wrong.

You're not the target user. Average users have no idea what you're talking about.

I think the target user is capable of going to chatgpt and saying "give me tattoo ideas"

By that logic, no one should build a startup that generates tattoos then?

Speak to average, non-technical users. You'd be surprised how many people have a very vague idea what ChatGPT is capable of. They aren't using it everyday like you and I. Relating this back to original comment, expecting them to know about effective prompting techniques, Stable Diffusion etc is unrealistic.

One of the reasons OpenAI offers APIs, is so you can build startups on top of their tech for average non-technical users.

> By that logic, no one should build a startup that generates tattoos then?

Is this like a law of the universe I'm not aware of, that you must be able to create a profitable startup that generates tattoo ideas?

> One of the reasons OpenAI offers APIs, is so you can build startups on top of their tech for average non-technical users.

A profitable product will make use of APIs to do something that a user couldn't do almost just as well by just prompting ChatGPT themselves.

It's funny that we went from "ChatGPT is going to unlock AGI and displace millions of workers" to "the only thing that came out of ChatGPT is a million of API wrappers that do nothing worthwhile at all" in like two years.

I mean it's basically the same thing as NFT/crypto grifters, just on a different tech stack. It's not about actually solving problems, it's about speculation to them.

Time to make a markov chain as a service startup...

> "the only thing that came out of ChatGPT is a million of API wrappers that do nothing worthwhile at all"

It’s because the base service does so much that people feel the wrappers aren’t needed, not because there are no real use cases for LLMs.

> Is this like a law of the universe I'm not aware of, that you must be able to create a profitable startup that generates tattoo ideas?

That's not what I said. Your question/argument was why build a wrapper when someone can go to ChatGPT and generate a tattoo. You can make that argument for any startup that wraps AI image models. If everyone followed that argument, there would be no startups in this space.

> A profitable product will make use of APIs to do something that a user couldn't do almost just as well by just prompting ChatGPT themselves

Exactly, average users won't know how to steer ChatGPT to generate high quality tattoos. In fact, OP is not even using an OpenAI model, they are using Flux which has no direct consumer interface (from the creators of Flux) and much higher quality image generation.

I wish OP success on his project.

I like the idea, though I couldn't get it to work. It just does the loading animation and then stops without generating anything or giving an error.

Same for me

Oh that's odd! Will have look into it, sometimes the API cuts out if it thinks it's NSFW.

prompt was "watercolor luck cat with cherry blossoms" hope it's not nsfw :)

Same for me. I'm guessing the traffic caused by being posted here made them go over their API limit... or it's just bused. idk.

edit... looked at network console. The API is returning a 504 gateway error.

Same for me as well

Looks like we hit a rate limit on the API which has caused the generations to stop working Looking to fix it now!

Hey HN!

Over the holiday break, I built TattooPRO as a learning project to get comfortable with the new Flux models for AI image generation.

TattooPRO, lets you create professional tattoo designs in seconds.

If you have an idea for a tattoo but can't find the right design to take into your tattoo artist, generate it with our AI. TattooPRO lets you create fantastic tattoo designs in 3 easy steps.

1. Enter your tattoo idea, and describe how you want your tattoo to look. It pays to be as descriptive as possible for the best result.

2. Choose the style for your tattoo. Do you prefer minimalistic or complex, perhaps something with more color? We have a wide range of styles you want for you to choose from

3. Get your tattoo design! When you're ready, hit create then wait a moment for the tattoo generator to present you with 3 unique concepts based on the idea and style that you've chosen. You can then download the design you like.

This has been a really fun project to get up to speed with AI image generation.

If you have time, give it a go, and let me know what you think!

-Ryan

> but can't find the right design to take into your tattoo artist

Isn't this what talking to the artist and having them design something for you is for? They have years of practice in listening to clients describe what they want and turning them into an image that works well as a tattoo. If you're cheap they also usually have a ton of pre-made designs known as "flash".

What kind of world do you live in where someone with "artist" in their job title can't work with you to create an image?

As someone with most of my body covered, two things I would say:

1) Sometimes I just don't really know what I want. I have an area X big and most of my body is in the style of Y. AI lets me iterate hundreds of design ideas quickly. My artist books a few months out minimum. It's hard to iterate.

2) References really streamline the process. I absolutely agree that my tattoo artist is an artist, so I don't go to them and say "I want this exact thing". However, most of my (and my artists) most enjoyable tattoos have been when I come with solid references and say "this, but put your touch on it" or "here's two ideas, can you combine them into something cool?".

I have a lot of my body covered too. In the beginning my tattoos were chosen/designed with a lot more care for ~meaning~ but nowadays all I care about is that the artist is "good" and has a style I like. Beyond that I don't need AI slop to figure out what to get on my body.

>all I care about is that the artist is "good" and has a style I like.

Same!

>Beyond that I don't need AI slop

I don't share the same hatred of AI. If something is cool looking, it's cool. I bring it to my artist as a reference and we work something out. I don't really care about the provenance.

very impressed with the images generated by the model. great work!

Thanks!

What mitigations do you employ against the high power draw of the GPUs in fossil-fueled datacentres that AI image generators use?

Who wants AI-generated Image tattooed on their body?

Robots, maybe?

Imagine meeting someone and your first thought is "Is that AI slop on their arm?"

Congrats on the launch and cool idea!

Some questions

- I noticed you limit to three free images, but incognito browsers effectively give you unlimited free images. Do you have a plan to prevent abuse, e.g. via fingerprinting? Did you notice people trying to abuse this? I'm more curious because I have a similar idea in a different space and noticed people running up my diffusion model meter.

- Not sure how long you've had this live, but do the paid accounts at least cover the cost of your infra/inferences?

- Why flux specifically? Was it an optimization on cost? latency? quality? ease? I see you're on GCP, why not keep it inside GCP all the way and use Gemini?

But who guarantees me that a tattoo artist would actually be able to draw these suggestions? Isn't that usually something you have to work out with the artist beforehand?

Honestly seems a bit tonedeaf considering doing tattoos has been one of the last refuges for artists looking to get paid for their art.

You can still make money on furry art

I mean, you should always be looking at what your tattoo artist can do before you get a tattoo. Don't print off a color cartoon picture from here and go to your local black-and-white realism artist.

I appreciate that you made it a one-time payment rather than a subscription service.

If you can find a partner that would allow your customers to order the design as one of those temporary tattoo stickers for kids, it would be an awesome product.

Prodigi are a company I've used for print on demand stuff before (a snapshot of the sun in different spectra given a specific time like a birth), though I've only done art prints they do have temporary tattoos.

https://www.prodigi.com/

UPDATE: API looks like it's back up! Hopefully it'll last longer next time.

Didn't expect the hug of death with this haha

UPDATE: Looks like we hit a rate limit on the API which has caused the generations to stop working

Looking to fix it now!

Very cool! What kind of prompts/methods have you found effective for getting clear backgrounds?

It was surprisingly simple to get clear background. I'm just adding this to the end of the prompt "The design should not be cut off and it should have white space around it."

wow 8000 tattoos were generated since this post went live

Finally a good use of AI art.

Although I never understood tattoos. High cost, painful, and (in some circles) culturally looked down on. I just don't get the motivation.

Thanks, I get that! I've only got 1 myself but it took far too many years to pull the trigger on that haha

There are cultures where tattoos are looked up on and symbolize achievement and status

> I just don't get the motivation.

It's "just" art. Some folks like showing off their style via their homes, some their cars, some their bodies.

Which makes an AI generated tattoo even more perplexing.

Why does it? Tattoo's are about your own personal style and expressing yourself.

It doesn't matter if it's AI generated or not. Also these can be a great start to take to a real artist and then they can work off that if you want.

Everybody is different, but for the people I know with tattoos picking an artist and having them create a design based on their input was a big part of getting their tattoos.

At the end of the day, if that's not a big part of it for other people then more power to them. AI generated at least ensures some level of image quality.

To me, it has all the charm of a brick.

>based on their input was a big part of getting their tattoos.

For me, as someone with ~70% of my body covered, I would have no problem using AI-generated images as part of that input.

I'm not just going to get whatever the model spits out as a 1:1 tattoo. But when I have a hard to describe idea, having some references in my pocket to pull out and say "something sort of like this" is helpful.

I tend to agree, which is probably why I don't have any tattoos, but my friends that do have them make me suspect that they may care about different stuff than us.

Most of the tattoos that I've seen are frankly uninspired. Dragons, flowers, skulls, knives. Some of the more unique ones are album artwork, but how creative is that really? Yet nobody seems bothered by this, they all show them off proudly! (There are some exceptions... My cousin had a full sleeve of Freddy Krueger making the pussy eating symbol with his blade fingers that was probably the crown jewel of some artists portfolio. An odd choice to be sure, but an impressive piece of work.)

I've come to believe that the draw here is the act of personalizing one's own body, not the quality of the art.

I have a friend who has this on her back: https://ocean.si.edu/through-time/evolution/darwins-evolutio...

she's really into science and evolution in particular.

I don't think it's particularly pretty, but probably unique.

Reply on news.ycombinator.com