.jpeg)
From the comments I’ve read, everyone seems to be of the opinion that the Christmas and Year of the Snake contests weren’t great choices for feeding a daily challenge board. These contests made the daily entry overly samey due to the relatively narrow themes. The default “World Morph” doesn’t really do much better (unless you’re doing stupid puns, is “stuff made of diamonds” really that different from “stuff made of cheese”?). I have a possible solution of a contest theme that would avoid these problems (it's actually something I proposed before, but figured I'd restate publicly) .
I propose, if contests are still a thing next year, that January 2026’s contest should be a Public Domain Day Contest. The contest’s unifying theme, and rule, would be that all submitted LoRAs must be made exclusively from public domain and/or Creative Commons (0 and By) content AND release their training data. This would make for a very broad range of possibly entries (the public domain composes all but a relatively very small portion of history) while keeping out most things management has expressed they don’t want contests being filled with (random anime girl art: Short of drawing it yourself and releasing the rights, or some early Japanese visual media being rediscovered it’s not doable).
Benefits
By requiring release of training data, the contest accomplishes something no other can: Let users see exactly the kind of data that goes into making a LoRA. It's one thing to read what kind of things go into a data set, but it's another to actually see it.
The other, possibly even bigger benefit, is that the possible items for such a contest are quite broad, but not excessively so as to deprive it of a true theme, freeing us of weeks of sweaters and scales. Below are just a sample of possible LoRA types.
Also 2026 is Year of the Horse. We all know a Year of the Horse Contest can't possibly end well.
Public Domain characters
The obvious one. There’s plenty of famous characters with consistent visual depictions in the public domain, including those from early cartoons and comics as well as characters from written media who had ample illustrations and book covers (such as Sherlock Holmes and Watson). Comics in particular often have a few more decades in the public domain since quite a few never had their copyright renewed back when that was a requirement in the US (they were both considered disposable and often by companies that didn’t last long).
Artstyles
Another big one. Plenty of famous artists and art movements are far enough in the past to have become public domain. These can be as ancient as general styles like cave paintings, Egyptian mosiacs, and Greek pottery to pulp magazines, as well as specific historical artists both well known (e.g., Leonardo, Frans Hals), more obscure (e.g., Harmen Steenwijck) and relatively modern (e.g., John Singer Sargent).
Famous Locations
As shown by my Ankor Wat LoRA, you can capture the unique feel of real world places both natural and manmade, which are well represented in public domain (and CC) images. Besides just being a pretty backdrop, adding atypical tags to the prompt or blending it with other location LoRAs will get you some fantastical locations.
Real Objects (and Clothing)
Modern models can do a pretty decent job of rendering objects, but struggle with very specific ones (both “this make and model” and some arbitrary things it, strangely, doesn’t know). Thanks to works of US government being public domain and many European state agencies releasing photos as CC-By, public works and especially miliaria are actually quite well represented.
Sports Poses/Equipment
A lot of sporting event pictures have been released into public domain, CC-0, or CC-By. Could absolutely make some LoRAs off this.
What would Civitai need for such a contest?
Civitai would only need two very, very minor features for this to have maximum viability. 1: It would need to adjust the wording on data set releases to explicitly OK public domain/CC 2: Allow attaching a txt file with LoRA and training data for proper CC-By crediting (you can already do this attribution in the model description, but it's unwieldy for larger datasets). An optional extra that could enhance such a contest would be a system that lets the public training data (at least, within reasonable file size) persist after 30 days if it was accepted into this contest
Conclusion
As you can see, a public domain contest would have a lot more to offer than most of the themes that have been done so far.