[edit: this certainly was a great success and a mistake, lol; besides generating wide shots as they were candies, at positive strenght it adds a classical painting effect, a sort of human absence and a "boring effect".
At negative strenght instead it adds... Refraction? Brightness? Colors? Interesting subjects? Asking for a chess game at -1 strenght I got a werewolf running from an explosion, surrounded by chess pieces: badass!]
Trained on 150 images of paintings of the surrealist current (think about Magritte, Dalí, de Chirico, ...), plus some stock photo and some danbooru image. This is probably a great error, made exclusively for the lulz.
I did NOT use AI generated images, nor a lot of modern surrealist images (as flying whales or whatever). If you want flying animals you digit "in midair", duh!
Some tag I used that usually would not be tried: hidden face, face revealed, inversion, perspective, human shape, scenery, human parts, weird, ...
Cross your fingers and let's see what comes out
Description
Same LoRA, 5th epoch. Being a concept I don't really know what is impacted upon, so let's experiment...
The psychedelic effect is A LOT less pronounced, as the painting effect in positive. It's probably the most versatile of the set, I think, if you don't like messing all the colors.
FAQ
Comments (7)
непон
I do not understand?
some russians think everyone understands russian language or they just simply dont care and type in in their language, they dont give a fuck about anyone else
@HEYHOI мне должно быть не похуй? если захочу донести мысль, напишу на понятном автору языке. Дело не в неуважении или рашизме. Я скачал модель, протестировал и не понял в чём её суть. Захотел и написал комментарий какой посчитал нужным
@yuduz367 ah, google translate managed to help with this one, I appreciate the feedback.
Ok, to explain: I did this back when I still didn't know anything about training LoRAs, so I used "concept": the style wiil be that of the base checkpoint.
This one should have helped recreating paintings of the surreal artistic movement, and does poorly even that, BUT!
If you push the weight in negative, it will give you a more extreme perspective, usually from below or from above, and colors more contrasting and saturated. Useful to generate characters running toward the viewer, for example. I haven't still checked how it works with the new generator.
@Argiop33 I tested this model on the model you use, and on my regular model. It seems it rarely produces the results you're talking about. At least, the result becomes chaotic and interesting if it's a simple prompt consisting of no more than one or two concepts.
@yuduz367 you are correct, sadly. As I wrote somewhere in the notes or description, that sort-of-wildcard effect disappears with extremely detailed prompts and/or NSFW generations, and probably is more consistent with 'noisy' base models. Being my second attempt, back then I considered it enoug. The version with a red werewolf on the cover is the strongest one, more epochs. And, as I said, I didn't test it again after all the changes to the generator some week ago (but other LoRAs became barely usable, so...)