
Trigger word: 'level'
Recommended settings: turn the CLIP strength up a little from the minimum and reduce the model strength a little from the maximum, because the dataset is not particularly diverse. I haven't thoroughly tested or anything so you may benefit from experimenting here.
This LoRA is intended to generate images with a 'filtered pixel art 2D platformer' appearance. Sometimes it renders raw pixel edges instead, though - presumably the base model's dataset contains a lot more unfiltered images in this style than filtered, so it's drawing inferences from there.
Quirks:
-
Due to the limited dataset, if it has no idea what to do with your prompt, it may draw an image in the correct pixel style but without making a platformer-like image with a floor
-
Due to the limited dataset, may transplant inappropriate floors into images where it makes no sense
-
Due to the limited dataset, biases towards making images with flat floors
I made many versions of this LoRA with synthetic data added to fix the narrow dataset but, in the end, the original version was the best. Maybe I'll fix it at some point.
描述:
训练词语: level
名称: level-a.safetensors
大小 (KB): 167939
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success