Show HN: Use SDXL fine tunes to make pro-quality maternity photoshoots at home(maternityphotoshoot.com)
maternityphotoshoot.com
Show HN: Use SDXL fine tunes to make pro-quality maternity photoshoots at home
https://www.maternityphotoshoot.com/
6 comments
Can you share some of your process, like is a full custom sdxl finetune really required for high quality results ?
Did you experiment with IPA face adapters or other techniques ?
I hope this works out for you, its a great niche good luck :)
Thanks! I suspect I'll throw this all away for SD3, but –
It's a proper fine tune (LoRA, not "full"). I experimented with IPA faceID and a few others but got poor results. Especially when training characters that don't look like celebrities. It's really obvious when you train it on yourself vs a stranger. And in this case the character is the customer, so that matters. As an aside this whole space of "custom fine tunes" will eventually get hurt by people promoting IPA products, which perform worse, but are 100x cheaper than real fine tunes. That is, until IPA-style tricks get better. (So, 3 months from now?)
Here's what I learned after lots of experimenting: stick to koyha's defaults. I wasted a lot of time on custom pipelines to normalize training images. I also wasted a lot of time fixing Replicate's SDXL cog, which I now know is more of a proof of concept than something people are supposed to actually use.
Don't use reg images (not necessary in this case), use a "celebrity doppelgänger" tool to train an existing token instead of a rare token, and, importantly, give folks a tool for spot correction (inpainting) to e.g. fix hands. Pretty much every image has some AI artifact but you can cycle through a few corrections to fix it.
Not sure if you're building in this space but I'll add: try paying for a few of the headshot generators out there. They're a bit disappointing. I assumed they had all "cracked it" so I kept working until I had a decent process, and the result is arguably better than the current leading products (which mostly use upscaled & face-corrected SD1.5).
On the market and product – this landing page is converting paid traffic very poorly and has had zero sales, so it's impossible to predict which ideas resonate, even if the tech is interesting!
It's a proper fine tune (LoRA, not "full"). I experimented with IPA faceID and a few others but got poor results. Especially when training characters that don't look like celebrities. It's really obvious when you train it on yourself vs a stranger. And in this case the character is the customer, so that matters. As an aside this whole space of "custom fine tunes" will eventually get hurt by people promoting IPA products, which perform worse, but are 100x cheaper than real fine tunes. That is, until IPA-style tricks get better. (So, 3 months from now?)
Here's what I learned after lots of experimenting: stick to koyha's defaults. I wasted a lot of time on custom pipelines to normalize training images. I also wasted a lot of time fixing Replicate's SDXL cog, which I now know is more of a proof of concept than something people are supposed to actually use.
Don't use reg images (not necessary in this case), use a "celebrity doppelgänger" tool to train an existing token instead of a rare token, and, importantly, give folks a tool for spot correction (inpainting) to e.g. fix hands. Pretty much every image has some AI artifact but you can cycle through a few corrections to fix it.
Not sure if you're building in this space but I'll add: try paying for a few of the headshot generators out there. They're a bit disappointing. I assumed they had all "cracked it" so I kept working until I had a decent process, and the result is arguably better than the current leading products (which mostly use upscaled & face-corrected SD1.5).
On the market and product – this landing page is converting paid traffic very poorly and has had zero sales, so it's impossible to predict which ideas resonate, even if the tech is interesting!
Such a great find! I wonder, how long did it take you to go from idea to launch? Did you worry about the implementation or are you going to do it manually at first? Anyways, great idea, really hope it works out!
about 7 weeks from idea to launch –
Idea: jan 6th, after a long road trip listening to danny postma and pieter levels interviews
I bought a cheap but relevant domain, put up an AI-generated landing page and bought some ad traffic to make sure I could get a decent CTR and advertise against the keywords without issue, and to confirm ahrefs is correct about it being a high-volume phrase. (Google is particular about health-related keywords.)
that worked, got a CTR of 10% and good volume, so i paid $1k to acquire the "real" domain, hired 5 models from Craigslist to experiment with different training techniques and to generate real examples for the landing page, spent about a month automating a pipeline for training, then, for the landing page, paid $400 for a human copywriter (which I really, really recommend – it turned out so much better than AI copy), $1500 for the design and implementation on Webflow, and a few hundred for misc expenses.
For implementation, signup, payment, image upload and order submission is all implemented. Using phoenix liveview which I love. But I am waiting for actual purchases to connect the training pipeline to the webserver.
launch: feb 27th with paid traffic via google, instagram and reddit (all performing poorly!)
Idea: jan 6th, after a long road trip listening to danny postma and pieter levels interviews
I bought a cheap but relevant domain, put up an AI-generated landing page and bought some ad traffic to make sure I could get a decent CTR and advertise against the keywords without issue, and to confirm ahrefs is correct about it being a high-volume phrase. (Google is particular about health-related keywords.)
that worked, got a CTR of 10% and good volume, so i paid $1k to acquire the "real" domain, hired 5 models from Craigslist to experiment with different training techniques and to generate real examples for the landing page, spent about a month automating a pipeline for training, then, for the landing page, paid $400 for a human copywriter (which I really, really recommend – it turned out so much better than AI copy), $1500 for the design and implementation on Webflow, and a few hundred for misc expenses.
For implementation, signup, payment, image upload and order submission is all implemented. Using phoenix liveview which I love. But I am waiting for actual purchases to connect the training pipeline to the webserver.
launch: feb 27th with paid traffic via google, instagram and reddit (all performing poorly!)
Fantastic! I think it’s a phenomenal idea and probably somewhat underpriced.
The one thing I would do is include the “before” selfies before the “after“ images.
The one thing I would do is include the “before” selfies before the “after“ images.
Thanks tomcam! This is probably the weirdest project I’ve launched so far.
I can share that the strategy here is to rank for “maternity photoshoot.” Obviously. But it will take a while to get backlinks and rank.
According to ahrefs the phrase has great volume (25k/mo in the US) and low competition (difficulty of 2, on a scale of 0-100). Which is an opportunity I haven’t seen before.
The low price is to incentivize initial sales and prove to people that it actually works.
This essentially follows Danny postma’s playbook, here: https://twitter.com/dannypostmaa/status/1646368426246680579
The alternative (a physical photoshoot) is expensive, and a luxury most moms can’t even consider, so it would be cool if the idea works out!
I can share that the strategy here is to rank for “maternity photoshoot.” Obviously. But it will take a while to get backlinks and rank.
According to ahrefs the phrase has great volume (25k/mo in the US) and low competition (difficulty of 2, on a scale of 0-100). Which is an opportunity I haven’t seen before.
The low price is to incentivize initial sales and prove to people that it actually works.
This essentially follows Danny postma’s playbook, here: https://twitter.com/dannypostmaa/status/1646368426246680579
The alternative (a physical photoshoot) is expensive, and a luxury most moms can’t even consider, so it would be cool if the idea works out!