Ben Brown noticed that my bot @5point9billion made him a personalised animated GIF when it tweeted him yesterday (on the occasion of light that left Earth as he was born, right at that moment passing the star Iota Pegasi, a little over 38 light years away). And he was curious about how it did that. So:
There’s a previous write-up about @5point9billion here. From that post:
My new bot is called @5point9billion which is the number of miles that light travels in a year. The idea is that you follow it, tweet it the date of your birth (e.g. here’s my starter tweet ), and then it lets you know whenever you reach Aldebaran or wherever.
You get tweets monthly, and then weekly, and for the last couple of days… and then you pass the star. It feels neat, don’t ask me why.
Since that write-up, I’ve also added a website to the bot. In addition to getting the realtime notifications on Twitter, you can sign in on the site and see what stars you’ve already reached.
Check this out: There’s also a public view, with an animation. This is a 3D animated map of all the star systems we can see from Earth, within 100 light years. It sits there and rotated. You can type in your date of birth, and it’ll show you what stars you’ve already reached.
I made this public view as a "kiosk" mode when @5point9billion was exhibiting at the Art of Bots show earlier this month. The stars were laid out on the floor, fanning out from the Sun which was right by the kiosk. Here’s a photo. It was good fun to walk out from the Sun till you find the star you’ve just passed. And then to walk out to about 80 light years and think, hey, most people die around this point, and look at the stars falling just further from you and think, hey, I probably won’t reach those. Huh.
And doesn’t it look kinda the same as the personalised star map that the bot made for Ben? Yup.
Making animated GIFs
I knew I wanted to tweet out personalised, animated star maps, whenever a bot follower passed a star (there are over 500 followers, and between 2 and 5 of them pass a star each day).
Routes I considered but discarded pretty fast:
- Generating the star maps offline. For sketching on my Mac, I use a Python drawing package called PlotDevice — this is what I used to make the first quick-and-dirty star map . I don’t like generating graphics offline because I want the ability to tweak and change my mind
This is the rendering pipeline I settled on:
- I already have queues and asynchronous processing on my website. The website is all Python because that’s my preferred language, and I have a my own Twitter bot framework that I’m gradually building up (this is a whole other story)
- When a user passes a star, the machine responsible for that task adds a tweet to the send queue, and flags it for requiring media
- At the appropriate time, the queue runner loads the animation page using PhantomJS which is a web browser that can run headless on the server. It’s possible to drive Phantom from Python using Selenium
- Because the animation is created on demand, and generated just for this tweet, it can include personalised information like today’s date, and the name of the user
- Using Phantom, each frame of the animation is generated by calling step(), capturing as a screen shot (as a PNG) to an in-memory buffer, and then down-sampling to half its original dimensions (this makes the lines sharper)
- Using images2gif (this is the Python 3 version of the library), the frames are assembled into an animated GIF, and saved as a temporary file
- The GIF is optimised by shelling out to gifsicle , a command-line tool for that purpose
- Finally, the media is uploaded to Twitter using Tweepy . Technically Twitter supports animated GIFs up to 5MB, but this is only available using a kind of chunked upload that Tweepy doesn’t yet support, so the GIFs have to come in under 3MB. Twitter returns a media ID, which the code associates with the queued tweet in my send queue, and that is posted when its time comes round. (The send queue ticks every 40 seconds, because Twitter rate limits.)
In summary, this is a rendering pipeline which:
- Fits my web-first approach… there’s no separate drawing package just for these animations, so debugging an image is as simple as opening a browser window
- Minimises the number of moving parts: I’ve added the ability to create images using Phantom but that’s it, there’s no separate drawing package or offline rendering
- Is agile: I can tweak and change last minute
What else am I using this for?
I prototyped this rendering pipeline with another Twitter bot, @tiny_gravity which just does a tiny particle simulation once every 4 hours. Sometimes it’s pretty.
This animation doesn’t use three.js for drawing, it uses processing.js , but the principle is the same. Again, the animation is just a webpage , so I can tweak the animated GIFs in the same way I tweak the rest of my website and bot behaviour. Here’s that animation as a tweet.
One of the things I’m most enjoying about having multiple projects is how they cross-pollinate.
My main side project right now is my bookshop-in-a-vending-machine called Machine Supply. Here it is at Campus , Google’s space for entrepreneurs in Shoreditch, London.
It tweets when it sells a book. Because of course it does.
The selection is changed over every Monday, and you’ll notice that each of the books has a card on the front (here’s a photo) because every book is recommended by a real human made of meat.
These cards and the shelf talkers (the label which says the item code and the price) are beautifully designed by my new friends at Common Works . But they’re a pain to produce: For layout, the templates are in InDesign (which I don’t have), then I have to send an Excel spreadsheet of the new stock over to Sam at Common Works, which he then puts into the template, and prints.
My new process comes straight out of the @5point9billion code. The browser is my layout tool.
So Sam moved from InDesign to the web, and here are this week’s shelf talkers as HTML. This is part of my admin site, I’ve temporarily turned off permission checking to this page so you can see. The template is automatically populated with details from the weekly planogram. (A planogram is the merchandising layout for a set of shelves or a store.)
And here’s the exact same page as a PDF . The pipeline is taken from @5point9billion: Phantom is used to grab the webpage, and this time render it to a PDF, complete with vector fonts and graphics. Because it’s a PDF, it’s super exact — which it needs to be to print right and fit neatly on the shelf edge.
It’s much quicker this way.
My rule for Machine Supply, as a side project, is that it should take the minimum of my time, never feel like an obligation, and I should be able to manage it on the hoof. As a hobby, it should be Default Alive .
So automation is helpful. I like that this mode of generating PDFs can be done without my laptop: I can do everything from my phone, and print wirelessly.
Anyway. You should follow @5point9billion! It’s fun, and you get a personalised animated GIF every time you pass a star, generated with the most ludicrous rendering pipeline ever.
转载本站任何文章请注明：转载至神刀安全网，谢谢神刀安全网 » How my Twitter bot makes personalised animated GIFs