Kamen Rider LoRA Announcement: Geats LoRA, Aesthetic Splitting, and the Final Three Rider LoRAs
With about 8 Kamen Rider LoRAs left, I feel like I should make a few announcements on the future of them. The first thing to note is that 7 of the 8 have already have their datasets made, which is just now the matters of cooking their LoRAs and then testing them for publication.
The eighth Kamen Rider LoRA is Geats which is a can of worms in of itself given it has about three variables: the helmet, the top portion, and the bottom portion, of which there are different sorts of combinations. What separates this from OOO is that there’s not any one set combination and that the design philosophy of the Geats Riders is that anyone can (theoretically) use all the Raise Buckles.
This leaves a conundrum because most of those Raise Buckles are just the same sort of shoulder strap and they give the person a weapon and the more standout ones are rare, with the upgrade ones like the Command Twin Buckles or even the upgrades to Boost being exclusive to the main Riders. So, when I do decide to make Geats’ dataset, I will do a compromise and do two LoRA sets:
The first will be the standard I’ve been doing for my other LoRAs, but in the style of OOO and Saber’s LoRA Dataset where only the fitting pair will get an Entry into the Dataset (and it will be their Fever versions to account for both top and bottom portions, though I’ll edit their eyes so that it’ll retain the Raise Buckle eyes), and with the Armed Buckles, the first non-main Rider to use it will be added, so it’s likely an earlier game Rider like Ginpen or Mary will be added than a late game Rider like Turbon or Brali.
The other will be more complex but something I learned about here, is to segment the Riders into their separate sections, while using all the Entry forms (something I won’t be adding to the first LoRA), or failing that, their more commonly seen form, as the profile shots to hopefully get a better representation of the lesser seen Riders.
However, I say when, since I’m going to wait for Geats to end. There may be new Riders or Forms waiting to jump out of nowhere either in the show’s endgame or even in the V-Cinemas and so I don’t want to go in making a LoRA half-cocked, so Geats will need to wait a bit. Same with the next Kamen Rider, which will be the first LoRA to become constantly updated as the year goes on with any luck.
However, now we go into the bigger announcement: As you might have seen, Kuuga and Agito have received a new update in which Kuuga’s forms and Agito’s Riders are split by tags. This is a new, experimental thing I’m doing for Kamen Rider LoRAs that I like to call “Aesthetic Splitting”.
As I brought up in the Gaim LoRA, there’s an issue with how Riders with different aesthetics will clash. To give you an example, I present Zangetsu:
Despite this being the same Rider having the same motif of a melon, you can tell that they have vastly different design aesthetics that make them separate enough. However, the LoRA can’t tell the difference beyond the fact that one wears their armor differently and has a bow because that’s what the dataset is saying. Same goes for Kuuga:
To the naked eye, we can tell there’s a difference between these Kuugas, but to a LoRA, it just sees two Red Kuugas, just one of them has more gold. However, the Grouped Kuuga LoRA’s dataset, however, can tell the LoRA that the Kuuga on the left is just Kuuga Mighty and the Kuuga on the left is Rising Mighty.
By splitting up the images via tags, they will increase the chances of that particular Rider or Form being called forward. This will also mean that I will be adding more images to the dataset, which means more data to feed on and more steps, which will most likely mean better results. Experimenting with this, however, I ran into something that is a clear indicator of its flaws:
For this Aesthetic Splitting to work, it requires there to be a large enough set within that aesthetic. You can see what I mean with the Agito LoRA, where despite requesting Gills or V1, I get a Rider that is more similar to G3 or Agito. I assume this also works for forms as well, so this may mean that only a select few Rider LoRAs will be getting this Aesthetic Split.
Maybe Ryuki (Survive can be doable if I pull from SICs but most likely not for Blank)
Kabuto (due to the Masked and Rider versions)
Wizard (Dragon Styles, Beast the only Rider who is left out aesthetically, is able to hold out on his own with the Mantles)
Gaim (Armored Riders and Genesis Riders)
Drive (Drive Driver, Mach Driver, and the Gunners. Might be a bit of a hassle though, so we’ll see)
Ghost (Boost/Deep)
Ex-Aid (Level 1)
Zi-O (Different aesthetics at play between Zi-O and Geiz alone, not even factoring in Woz, the Quartzers, and the Future Riders)
Zero-One (Same case with Build but there’s more data and consistent difference)
Saber (Bit iffy on this one but I think I can manage)
Meanwhile the LoRAs that would likely not qualify.
Faiz, Blade, Hibiki & Den-O (Not really enough variation in the data to justify the split)
Kiva (Ixa is the only other Rider whose aesthetic is conversely different from Kiva’s, but he’s the only one aside from Saga who is just in the middle)
Double (Same case with Accel)
OOO (Same case with Birth)
Fourze (Same case with Meteor)
Build (A different amount of Drivers at play, but not sure if there’s enough users or a consistent enough aesthetic difference)
Revice (Despite different drivers being used, the aesthetics are all over the place)
Even then, it’s something I won’t consider for all the items on the list that I’m considering, as the idea will be that they will have the unified LoRA and if there’s a demand to split them based off the different aesthetics, I will make a split version.
One final announcement before this article ends: Once all the Rider LoRAs are done (including Geats), I will be making an All Riders LoRA, counting the next Rider. More details will come on who will qualify for entry in the dataset but I think the indicator will exactly what it says: All Riders.
Afterwards, I will be formally splitting the Showa Rider LoRA into 2 new LoRAs outright without the Aesthetic Split feature: True Showa and Showa Homage. True Showa will feature all the Showa Riders from Ichigo to J (yes, I know Shin, ZO, and J are Heisei, but as far as most Rider featurettes are concerned, they qualify as Showa) and Showa Homage will feature all the Riders made post-J that either homage the old classics or are outright remakes (this is where G, Aqua, Century, the Goriders, and all the Riders from Amazons, Black Sun, and the three movie remakes of the OG Kamen Rider show will be at, along with other homage Riders I forgot to mention).
After those are complete (or if I’m waiting for Geats to be done), I’ll be back to making character LoRAs.
Well, that’s all from me today. Until the LoRAs are complete, peace out. Feel free to ask me questions and I’ll answer them to the best of my ability.