Toy Box Flux was trained entirely on AI-generated images. The concept was to blend the weights of an existing 3D LoRA by SECourses with my Coloring Book Flux LoRA. The results were incredibly cute, as Flux developed a unique style that I decided to pursue further. I selected approximately 71 synthetic images and used them to train a LoRA focused on the specific style I wanted to achieve.

71 images, 1 repeat, 25 epochs, 32 DIM / 32 ALPHA, 2,486 Steps

FOR THE BEST RESULTS USE EULER OR DEIS SAMPLER

This LoRA works best with objects and human subjects, animals are hit or miss, not enough data in the training images. Something interesting I found is that this actually helps increase the quality in realistic 3D renders of interiors as well.

In v2 I plan to train on more generated outputs intermixed with the pre-existing outputs to create a stronger adherence to this style.

Trigger keyword: 't0yb0x' 'simple toy design' 'detailed toy design'

Recommended strengths: 0.7 - 0.9

You should use t0yb0x to trigger the image generation.

You should use simple toy design to trigger the image generation.

You should use detailed toy design to trigger the image generation.

You should use 3D render to trigger the image generation.

Weights for this model are available in Safetensors format.

Download them in the Files & versions tab.