Psionix
(noun, colloquial, sai-on-iks)
Colloquial variation of psionics, refers to the use of internal mental, emotional, or willpower-based forces to achieve supernatural outcomes such as telekinesis, telepathy, teleportation, and extrasensory perception. Cool dudes spell words with a X.
A San Diego-based video game developer (unrelated).
This LoRA is an hommage to western comicbook art in the early 1990s and was generated from a curation of my own personal comic collection.
The 1990s was a crazy time for western comics. Coming off the heels of Chris Claremont's seminal X-Men run, the characters all had armored segments, enormous futurist guns, shoulder pads which outlived their real-life 1980s trend-death, and so very many pockets.
Psionix emulates both the comic-art style of the 1990s and the character designs. The men are hairy and burly, the women are buxom and hourglass-shaped, the costumes are bombastic and impractical... it's a real vibe. ๐
The art-style itself is marked by Ben-Day dots, giving it that fresh-off-the-press feel. The 1990s itself was an interesting time for western comic printing, as there was a shift to glossy paper media towards the middle of the decade and a reduction in the stylistic use of Ben-Day dots. As a result, when you lower the strength of the LoRA, you actually get a nice transition that reflects this style shift - moving from early 1990s up to early 2000s western comic art styles, with less pronounced Ben-Day dots and a smoother, more rounded art style to the characters while still preserving some of that character-concept bombastacism.
This is a strong LoRA - I recommend starting at 0.8 strength. Going up to 1 could be useful situationally, particularly if you want to get closer to that Silver-Age feel, but the style is kinda ecclectic in places, especially around it's build-a-bear futurist technology and sloppy background art, so choose wisely. Dropping down to 0.6 strength gives you a mid-90s gloss, and once you start going as low as 0.3-0.4 you're getting some heavy style bleeding weirdness that is fun to play with and smacks of the miniseries Marvels or Earth X, if you're familiar. Some weird comic-photorealism blending.
Hot Tip - when prompting for text (speech bubbles, flair text, etc.), putting it ALL IN CAPITALS seems to help the model since the comic-default fonts are all in caps.
One of the best things about this LoRA is that I avoided well-known comic characters in making it. This means that it skews away from making Superman designs when you prompt for a caped super-hero, and skews away from Spider-Man designs when you mention the word 'spider'. No Supermen or Spider-Men were used in the construction of this LoRA. ๐
It has very limited NSFW capacity, but lacks detail. It wasn't made to be NSFW, so you can use it for wide-shots which contibute to scene composition less than upclose pornographic titilation. However, plug in a NSFW LoRA and you can corrupt this bad boy to your heart's content. ๐
Due to the nature of the hand-drawn art style and the ecclectic gibberish that contibuted to some of its learning, it can struggle with anatomy. Luckily, this was true to the art style of the time. You can course correct by dropping the LoRA strength down or using prompts such as 'best hands, five fingers', etc.
I loved making this one, it was fun using a physical media to make it, and it's an art-style that was formative in my love of art itself. I lived and breathed 1990s X-Men, so great to see some of that style back in an accessible and customisable form.
Hope you guys have fun and get to channel some radness. ๐๐โโ๏ธ๐๐ค
Enjoy! ๐๐ค
Description
FAQ
Comments (10)
Ive been waiting for a good 90s comics style with klein forever now. Do you think you could do a 50s one next? Also can it do edit?
Glad I could fill the comic-shaped hole in your heart. ๐ซถ I love Golden Age comics, I would love to do a Golden Age LoRA, but I have a few other LoRAs on my list before I get there. Definitely wanting to do more comic-based LoRAs though, I'd love to do early 2000s and Silver Age before I get to Golden Age (sorry! ๐ ๐ค), and a Manga LoRA is on my to-do list too. I'm currently working on some custom nodes to assist comic-panel composition in ComfyUI, so rest-assured, comic art is a high priority here. Stay tuned! ๐๐
@DirkDigitallerย Doesn't seem to work that well with img2img, adding 90s comic sort of works, but it loves to populate with the comic dots.
@Starry_Eyesย Ben Day dots? Hmm, weird. I might play around with it too and let you know how I go. Haven't uaed F2K's I2I much, it's a good excuse. ๐๐
@DirkDigitallerย also, in a future version, besides img2img, try and preserve some of that 90s anatomy rather than just style, like extremely thin waists,
And for img2img, usually having a trigger word I think helps with loras.
@Starry_Eyesย The Qwen version preserves the 90s body-types and character-elements better than the F2K version, same dataset and captions, just an element of the base model and how it trains I'm afraid. ๐คทโโ๏ธ
@DirkDigitallerย i cant run qwen unfortunately. ive tried the gguf before, but every time I do comfyui manager bugs end up corrupting my install and I have to start over.
@Starry_Eyesย Ahh, that sucks, sorry. ๐ค I'll likely iterate a v2.0 on this at some point, maybe I'll crack the F2K training code and get better results next time. ๐๐ Otherwise, you could possibly run the Qwen checkpoints on Civit with the LoRA, if you feel like spending some buzz. ๐
@DirkDigitallerย thanks anyway, I actually like using these loras on my own physical sketches.
@DirkDigitallerย after playing around more, it works really well for text to image, i think it just needs a trigger word to work with img2img


















