Preview images do not use other LoRAs.<h1 id="commercial-use-of-this-model-is-not-allowed">❌💲 Commercial use of this model is not allowed</h1>A few of the dataset pictures to show what the character really looks like (questionable choice for a LoRA, I know)<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/9592bc3e-b8ae-4f98-8fcf-215be7525dcc/width=525/9592bc3e-b8ae-4f98-8fcf-215be7525dcc.jpeg" />The sample images were surprisingly bad, considering it's just a simple cartoony furry, so I'm not putting them into the showcase. I thought I've been unlucky with those just because I usually do Worms, but I guess sample images are just trash in general and you can never trust them... I may be spoiled by Illustrious and Noob, but I thought Pony would handle this character better than what I see. I didn't use elaborate tags for his markings so it's no wonder they come out wrong, but for some reason it struggles with ears and keeping the shirt yellow, among other things.<h3 id="context">CONTEXT</h3>Alik is the mascot of a 2000's computer game series for children, nearly all of which are edutainment. The games were developed by the Czech studio Silcom Multimedia. Unfortunately I haven't heard of these games having English versions. Alik would usually explain things that every child should know like colors, letters, numbers, etc. and tell them what to do. These games are well-remembered in Russia, but IDK about Czechia itself and other countries.I also found some Czech website using this character but I haven't looked into it due to language barriers.<h3 id="dataset">🖼DATASET</h3>Tagging is fully manual. The games' resolution is low (most are 640x480, one is 800x600), so some images were automatically upscaled. Some images contain other characters. I didn't tag facial expressions like "happy" or "sad" because there's little variety and he usually appears to be smiling or neutral. Background elements usually were NOT tagged.

Alik [PDXL-NoobXL]

Same dataset as PDXL V2. Trained against NoobAI epsilonPred v1.1 (we now have Chroma, Flux 2 and ZIT as default options, but STILL no Noob?)Unfortunately this version also suffers from deformed and extra ears, and I saw a lot of extra feet while testing a prompt with "jumping" in base Noob. Overalls have always been random, I don't care much for them at this point. Maybe training a LoRA with pictures from those games with such art quality was a fundamentally bad idea. I've only posted results from finetunes.

Files with SHA256: 0b37557c272d871a43736997da41f34daf109e99388ac14067ce708489d2d66d