Home
Starting Up
Case Studies DB
Products
Ideas DB
Vibe Coding Tools
Subscribe to IH+
Starting Up
Case Studies
Ideas DB
Products DB
Join
3
Likes
14
Comments
Roast: AI avatars in custom kid's books
by
Tom
https://tigerhousebooks.com
No mercy please... rip it to shreds!
The color schemes are a bit odd. I'd go with straight color and get rid of the gradient
It's very good otherwise!
Just wondering, what backend are you using for this? I'm building Evoke and getting dreambooth up soon, so think it would be a great backend for this app.
Also, have an active AI discord
In answer to your question, I was using Meru for tigerhousebooks training and running inferences on runpod. Pretty janky, not being able to do both in one place, but it was the first thing I got working and didn't have spare time to de-jank it!
Thanks!
So I've pretty much moved over to focusing 100% on https://artificial.pub now, which avoids some challenges I was having with visual continuity. It runs on replicate, which Just Works right now, but I would totally consider replacing that down the line once I'm not dealing with basic functional requirements that are still manual in a lot of places.
Interesting to hear! What basic functional requirements do you find are manual in a lot of places that replicate seems to solve?
Great question, possibly a poorly worded para by me. The things that are manual are around my own backend stuff, not related to AI/SD. The "lot of places" are all in my code.
Some of the things I haven't fully automated yet (in most cases it's partially automated):
In addition to all that, I have work to do on both marketing and UI, and it's all much more urgent than changing something that works :)
This is awesome! Any book samples?
Just came across it through talk of Dreambooth API. What is the interest in that?
Recently just made https://stablematic.com/ which is to help people run Stable Diffusion models in their browser (via Automatic1111) with the latest GPUs in one click.
Hey! I ran into a couple of hard problems (for me) with this one, especially around a) visual continuity and b) users' ideas about how they should use it.
You probably know what I mean by the former.. getting visual continuity with SD (not of the fine-tuned character, but of everything else) is beyond my skill*time quotient.
The surprising thing for the second point was that because it's printed as a one-off, almost everyone wanted to participate in generating the story. That means all sorts of complexity integrating LLMs as well, plus it means illustrations need to be redone each time. And I believe it would be a "homer simpson car" if I delivered what they ask for... we don't ask to direct our own movies or write our own books or design our own cars! And in fact the "write your own story" service already exists with every self-publishing service in the world. Super dumb idea! :D
But almost everyone said it, so I wasn't super comfortable.
TLDR: I asked myself "what if I make my challenges into constraints: what is good without continuity or story?" and I made https://artificial.pub instead :D
I used the code from https://photoshot.app as a starting point.
I'll check out stablematic.. right now I'm not exposing anything relevant, but in time that may change.
The text is way too small in certain portions, so it may just be better to have a privacy policy page rather than making it fine print at the bottom.
For the text, you're making it read like one of those funnel scam websites (not saying you're a scam obviously), so I would reduce the text and just make it look more like a standard corporate website
Also the background makes contrasting against text a little hard.
You should also be transparent with pricing. "Under $50" does not inspire confidence. Give customers the full exact price.
Also, are you having people making these through a UI, or are you making them yourself and sending it? Also, where do the users upload photos? Make that more clear.
I'm working on an AI tool as well called Evoke and we have a stable diffusion API, so if you ever go the route where you're building a UI for customers to make their own books, you can use Evoke or DM me on twitter here: https://twitter.com/TheRealEtch
Good luck!
Done. Reduced amount of text, increased text size, removed background image, increased contrast with background color, reduced weight of background pattern, rewritten/removed a lot of copy to give it a more serious tone.
I also rearranged the order so that the inpainted images are closer to the top, and moved the payment link to its own page with detailed instructions on the process.
Thanks again for your input.
Gold! Thank you so much, love it.
I don't know whether I'll ever open up design functions to customers. Makes it a lot higher touch.
That said, I want the training and inference to be as automatic as possible even when I'm the one doing it. Currently I'm running dreambooth with Meru's API and then doing inpainting by hand with automatic1111 (and static init image + mask) on runpod.
How far off is your dreambooth API?
Dreambooth API's ETA is not know at the moment, but we have a discord.
As for our stable diffusion API, the main benefit is that it's much faster than most on the market.
If you join, we'll be sure to let you know once we release!
Please re-route your eyeballs to my other project: https://www.indiehackers.com/post/roast-book-of-portraits-for-real-kids-by-dead-painters-32ef9ef961
Tiger House Books is on hold for now, all energy is going in to https://artificial.pub
Breaking news... I'm going to change the cover of the book to have the kid's face on it, and the first image will be a photo of two kids holding their own books. I'm hoping to make it instantly clear what the product is, and reduce the importance of AI a little further. The point is real kids faces in custom books.
Anyone... anyone... Bueller?