Code Lyoko was a French animated series that aired between 2004 and 2007 about a group of friends who <s>play </s><s>Where’s Waldo</s> fight an evil AI called X.A.N.A. that is capable of possessing inanimate objects in the real world. It featured separate 2D animated sequences for the “real” world (which this is based on) and 3D CGI animation for sequences within the computer world (which this does not include). Stylewise the series is still (well?) remembered for the use of large foreheads on most character designs. Probably not what she meant when she said "draw me like one of your French girls"Tag: C_L_F_H, For Humans: Black Eyes (V2 is trained on this, which is the only change for V2) Negative: other eye color tags <s>for generic characters. I don't know how force the eyes to be black dots within sclera for characters from a LoRA or tags.</s> V2 has black eyes baked in. Still want to avoid conflicting eye color tags and even use them as negative.I made this style because <s>fiveheads are better than one</s> after making my Himi LoRA, I wanted to generate her in this style because the mark on her head strongly resembles the main antagonist’s emblem, then saw nobody had made a style LoRA for Illustrious yet. Knowing <a target="_blank" rel="ugc" href="https://www.youtube.com/channel/UCn9zTBZVl6lu72s-6fXttZg">the episodes were officially on YouTube</a> and wanting to experiment more with styles (mainly how many images are actually needed) before I did something I’m also working on, I decided to give it a go.Made with 126 screenshots with minimal prep (a handful were cropped to exclude the main characters or, in one case, a naked Romulus and Remus from a reproduction of Lupa Capitolina in the background). The only cleaning for most images was to crop some distortion at the edge of images (couldn’t automate it because it differs across episodes). I deliberately avoided having the same character in the data too much, so character shots consistent primarily of (adult) background and secondary character to ensure variety (they actually have more complex face geometry than most of the main cast anyways!) and to check if style LoRAs made with screenshots are compatible with character LoRAs made with non-overlapping screenshots of the same series (Answer: Maybe at lower weights, but as it just hyper exaggerates the features). As many episodes really only have the recurring cast, finding enough took longer than I’d like and required checking all episodes, but it wasn’t that bad. (Note: This was made after I determined I was no longer committed to doing daily LoRAs, it just happens to have come out before my backlog expired rather than being a return. I'll likely give this a V2 eventually if I can figure out a solution to the eyes)

[IL] Code Lyoko (2D) Style

IL_Code_Lyoko_style_v2.safetensors

Code Lyoko was a French animated series that aired between 2004 and 2007 about a group of friends who <s>play </s><s>Where’s Waldo</s> fight an evil AI called X.A.N.A. that is capable of possessing inanimate objects in the real world. It featured separate 2D animated sequences for the “real” world (which this is based on) and 3D CGI animation for sequences within the computer world (which this does not include). Stylewise the series is still (well?) remembered for the use of large foreheads on most character designs. Probably not what she meant when she said "draw me like one of your French girls"Tag: C_L_F_H, For Humans: Black Eyes (V2 is trained on this, which is the only change for V2) Negative: other eye color tags <s>for generic characters. I don't know how force the eyes to be black dots within sclera for characters from a LoRA or tags.</s> V2 has black eyes baked in. Still want to avoid conflicting eye color tags and even use them as negative.I made this style because <s>fiveheads are better than one</s> after making my Himi LoRA, I wanted to generate her in this style because the mark on her head strongly resembles the main antagonist’s emblem, then saw nobody had made a style LoRA for Illustrious yet. Knowing <a target="_blank" rel="ugc" href="https://www.youtube.com/channel/UCn9zTBZVl6lu72s-6fXttZg">the episodes were officially on YouTube</a> and wanting to experiment more with styles (mainly how many images are actually needed) before I did something I’m also working on, I decided to give it a go.Made with 126 screenshots with minimal prep (a handful were cropped to exclude the main characters or, in one case, a naked Romulus and Remus from a reproduction of Lupa Capitolina in the background). The only cleaning for most images was to crop some distortion at the edge of images (couldn’t automate it because it differs across episodes). I deliberately avoided having the same character in the data too much, so character shots consistent primarily of (*****) background and secondary character to ensure variety (they actually have more complex face geometry than most of the main cast anyways!) and to check if style LoRAs made with screenshots are compatible with character LoRAs made with non-overlapping screenshots of the same series (Answer: Maybe at lower weights, but as it just hyper exaggerates the features). As many episodes really only have the recurring cast, finding enough took longer than I’d like and required checking all episodes, but it wasn’t that bad. (Note: This was made after I determined I was no longer committed to doing daily LoRAs, it just happens to have come out before my backlog expired rather than being a return. I'll likely give this a V2 eventually if I can figure out a solution to the eyes)

Files with SHA256: 3c6f1ba305c81bed02480434ca186c0ff683e682ecddf121dae057252d7d4d20