Pico Library uses artificial intelligence to write and read fun stories for your child. Powered by Apple's Neural Engine, all the magic happens right on your device. This means you're ready for story time wherever you are - at home, in a car, or on a plane
I'm pretty excited about the potential of small language models. Advances in language models, coupled with increasingly powerful mobile devices, have made it possible to carry language models with billions of parameters in our pockets. While I was on paternity leave, I started building Pico Library to make storytime more engaging for both me and my child.
As a new father, I've come to deeply appreciate the importance of reading to my child. Pico Library uses a language model to write stories, and it can leverage Apple's on-device Personal Voice feature to read them aloud. Everything is powered by on-device computing, meaning no internet connection is required.
I'm happy to chat about the technology in Pico Libraryβwhat I'm excited about and what I found surprising. I'm also curious to hear what others would build with similar technology.
Report
Hi @jenhan_tao, do you think adding AI-generated images throughout the text would make it more kid-friendly?
@pat_miolan I think adding some AI-generated illustrations would definitely make the stories more appealing to kids! There's AI image generators that can run on iOS devices, and so I think the key trick is to figure out which portions of the story to illustrate.
Report
@jenhan_tao That would be a great integration then. Good luck!
Report
I like this. I love reading short novels. But I prefer someone reading that. This is absolutely what I need.
@sentry_co I'm using Apple's Personal Voice feature: https://support.apple.com/en-us/...
To create a Personal Voice, you record yourself reading phrases for a few minutes and then you can use your own voice clone to read stories. My own voice clone is a bit stiff, but it definitely sounds more human than the default system wide default voice.
I really liked that Personal Voice is created and deployed on-device and so there's no concern about your voice clone being misused
@jenhan_tao Yeah. That's a really cool use case! You can create realistic voices now with AI with just a few seconds of recording. I forgot about that. That's really cool. And a bit creepy. You can basically resurrect dead relatives RIP gets a new meaning. π What are your plans to take this another level? are you going to build out the app?
@sentry_co There's definitely more work that can be done on the writing model side so that it can: tune the vocabulary for the child's age better, write new stories, and perhaps write sequels to stories.
I think my initial instincts for what to do next leans towards model development tasks because that's more related to my day job. I think to take this to this to the next level, I need to get some feedback from users, and focus on building a compelling tool for parents.
Beyond this app, I'm pretty excited about ML models and AI that fit in our pocket, and I may explore other applications/problems!
@jenhan_tao Yeh. Im super excited about Local models too. Currently I want to start using the local model Ollama , and use its coding model instead of GitHub copilot chat. Ollama also has support for swift. So that might be relevant for you too. I think the future might me local models. And then you just patch them with small updates. Avoids the whole send data back to the mothership issue etc.
@sentry_co I definitely recommend trying out ollama. think ollama also uses quantized/compressed models. Itβs pretty surprising to me you can compress models to a fraction of the original size yet still preserve most of the utility. Relying on a compute expensive mothership also means higher costs - I saw other AI story generators that wanted a monthly fee to (probably) cover their compute costs.
This is a fantastic idea, especially for parents on the go. How customizable are the stories generated by the AI? Can the content be tailored to specific interests or learning levels of children?
@kyrylosilin thanks for the positive feedback, and I really hope this is helpful for at least one other parent!
Right now, there's three customization factors that can be edited in settings.
- Name - the main character's name can be specified
- Age - lower values encourage simpler language and higher values encourage more complex language.
- Favorites - things/topics your child likes will be incorporated
Surprisingly, vocabulary level is tougher to control than favorites.
The "Customize Story" toggle can be used to turn customization on/off.
I'm actively working on finetuning the underlying model so that stories can be better customized - stay tuned for updates!
Report
As a father, I'm glad someone is focusing on the importance of this area of parenting, and incorporating AI into everyday education is a way for technology to bring extraordinary results to ordinary routines.
Thank you.
@horus_he Thanks Horus! I'm hopeful that some parents will find Pico Library helpful. I think on device computing can lower the cost of deploying AI and ultimately make AI technologies more broadly accessible
@eileen_lim Thanks Eileen! I was super nervous about my first flight with my kid, but I'm definitely curious to see if Pico Library can keep them entertained on future flights
Pico Library
Pico Library
Pico Library
DiffSense
Pico Library
DiffSense
Pico Library
DiffSense
Pico Library
Telebugs
Pico Library
Pico Library
Just Scroll
Pico Library
Pico Library