Where on Earth Is Carmen Sandiego? was an American animated edutainment series that ran from 1994 to 1998. It was a spinoff of the 1985 edutainment video game series Carmen Sandiego . Everything about this series oozes 90s, with even its educational content managing to be extremely dated (both because of new discoveries, and geopolitical upheaval, but also having relatively weak scholarship that's very obvious in the easy research of the internet era). (Now that the same company owns both IPs, at least in animation, I wonder if we’ll ever get Inspector Gadget vs. Carmen Sandiego where Brain/Penny and Carmen try to out-do eachother in disguises.)<a target="_blank" rel="ugc" href="https://civarchive.com/articles/18582/20-loras-and-images-the-credit-card-companies-dont-want-you-to-see">To see all my LoRAs and all the example images, please (temporarily) disable X and XXX visibility, otherwise a lot will be hidden at the whims of credit card companies</a>. You can help support making models like this <a target="_blank" rel="ugc" href="https://civarchive.com/articles/9229/how-to-give-a-model-creator-50-buzz-at-no-extra-cost-possibly-negative-to-you">by posting stuff you've made with it to the gallery using the "add post" button</a>, that helps me earn back the training cost at no cost to you (and I just plain old like seeing what people do with the LoRAs I make).Trigger: W_I_C_S In data a lot and works well: dark, nightMade with 104 screenshots I hastily tagged, relying mostly on auto-label beyond obvious corrections (mostly male vs. female) and images that barely got anything. The uploads appear to be from a broadcaster’s tape instead of masters, so I kept care to avoid obvious artifacts in screencapping. This show has a smaller human cast than you’d think (at least in the episodes uploaded) making it actually quite difficult to get non-Ivy, non-Zak, non-Carmen, non-Generic Henchman humans, even from nameless characters, leading to lots of generic background characters being in the dataset and a high number of no humans pics (27).I saw that <a target="_blank" rel="ugc" href="https://civarchive.com/models/557912?modelVersionId=2018197">a Carmen Sandiego LoRA</a> trained on this show had a nice inherent style. Since I’d already wanted to experiment with using primarily curated auto-labeled screenshots (instead of full handtagging) to make a style, was trying to narrow down just how many screenshots are needed for a style anyways, and <a target="_blank" rel="ugc" href="https://www.youtube.com/channel/UC69mJ70uL0OzCp-NFzQz5LQ">a chunk of this series was officially free on YouTube</a> I decided to give it a try. I don't think the results are as close as I'd like, but I do think it makes something nice regardlessV2: I realized after making this that the official upload I used for my training data was stretched to 16:9 and that's why everything was so weirdly wide. Ran a <a target="_blank" rel="ugc" href="https://civarchive.com/articles/19677/linux-mass-crop-or-resize-images-with-no-extra-programs-using-command-line">mogrify on the training data to unstretch it</a> and retrained with no other changes.

Files with SHA256: ef9c6a400a6cd83b4275567bde961539fb3d9314d2aa06b7094e6f19cf16296f