Toon-Tacular Qwen

Trained on 70 curated images, the Toon-Tacular Qwen LoRA breathes character and expression into your generated images. The style is reminiscent of mid-to-late 90s and early aughts cartoons. The dataset was regularized by using an edit model to upscale and unify the style to be consistent. The goal was to give all the aesthetic with less of the degradation/compression.

The LoRA was trained with the fp16 version of Qwen Image 2512, and tested with the same model, it's far from perfect but generally maintains the style consistently. This LoRA currently has weaknesses with overly busy backgrounds, smaller faces and some anatomy. The trigger word is t00n but it's not necessary to use it, simply including words like animation or cartoon triggers the style. Use an LLM and be strategic in your prompting for the best results, this isn't a one shot type of LoRA.

The first image in the gallery will contain a workflow that I used to generate the image. You don't have to use it but I'm including the embedded workflow in the image for completeness. You're welcome to modify to fit your use case. If it doesn't work for you then please skip it, I will not be offering support beyond sharing it.

Trained with ai-toolkit and tested in Comfy UI.

Trigger Word: t00n
Recommended Strength: 0.7-0.9
Recommended Sampler/Scheduler: Euler/Beta

renderartist.com