I also updated the script that sent the tweets to post in Discord when the queue ran low. When prompted, via a !review command, it would pick a skin, send a message with the skin's screenshot attached, and wait for a y/ n response. I figured a better UI - and the ability to review from my phone - would help, so I rewrote the review script as a Discord bot. I would often fall behind and feel guilty as I went about my day knowing that I was behind on reviewing skins. It turns out, it was rather tedious to keep up with the bot's appetite for reviewed skins. I also added some code so that if it noticed that the list of approved skins was getting short, it would email me a reminder to review some skins. So, I adjusted the review script to read/write the text files from an S3 bucket and moved the Python script to a server where I had an account. In order to send the tweets automatically, I needed to have the bot running on a server since my laptop is often closed/off. The script would then record the filename in one of two text files.

I wrote a script which would pick a random skin, open its image in Preview, and then prompt me to approve or reject it. This meant I couldn't blindly tweet a random skin, I needed a review process. The biggest challenge with running the bot was that, because I was trying to collect as many skins as possible, many of the ones I had were of very poor quality or were 'not safe for work'. This ensured that even when upscaled using the smoothing approach, the pixels were relatively distinct. Next, I manually upscaled the image to twice its original size by doubling each pixel. The fact that the image (technically) had transparency forced Twitter to keep the images as a lossless. First, I set one pixel's alpha channel to 254. I was able to work around these two issues by adding a step to my script which would pre-process the image before uploading.

Additionally, the scaling used to present the images at a large size assumed all images were photos and smooth out the pixels as it scaled, rather than keeping the pixels distinct and crisp. This meant that the pixel art images I was uploading were getting presented in a lossy format optimized for photos and, as a result, looked pretty awful. I soon realized that Twitter was reencoding my. It was a manual process and I tried to run it a couple times a day from my laptop. I loved the idea, so I registered an account and hacked together a Python script to tweet an image with a link to that skin loaded in the bowser. After writing a script combing Puppeteer and Webamp to automate taking screenshots of Winamp skins, on Twitter suggested I create a Twitter bot.

2017 best winamp skin