Concept of the meme Foot Burger.

This is a low quality LoRA made on Ostris AI Toolkit just for a test of my first Z-Image LoRA training. It was trained in 512p to be faster and see if the parameters are good enough and if its worth waiting a longer training time in high resolution as my hardware is not good enough.

From my tests Z-Image got the concept pretty well but I guess my LoRA didnt trained that well because of the parameters.. so regenerating and looking for better seed are still needed. I believe it can be much better. I may or may not try it again, testing with limited hardware speed is complicated. I'm still searching and checking other people parameters who had a good success with it to try.

Use this prompt and modify it as you wish according to the instructions below, but keep the same structure, see prompts in the examples for a better idea:

In a POV view, a man/woman's hand is holding a hamburger bun with her feet as the filling.

Specifying "hands" or "hand" the model (on my tests) most of the time will understand that you want one hand or two hands holding the hamburger bun as I included it on dataset.
Same for "both feet (on dataset only "feet" was used for both so use it first before written "both") or "left foot" (for one foot or two feet on the hamburger bun).
And specifying hand characteristics, such as male/female, skin color, nail color and length, works, according to my tests.

This lora does NOT work on Z-Image Turbo, use Z-Image Base. I did some tests on Turbo and the outputs are bad. According to other people training on Z-Base using Prodigy will work better on Z-Turbo but in my case it failed.
Test different seeds and resolutions if the output are not good. I recommend testing generating three times in these resolutions 1248x832, 1562x864 and 2048x1152, some resolutions gave better results in specific seed/prompt.

Examples are made with Z-image Base Q8 gguf, 30 steps, CFG 4, sampler res_multistep and scheduler simple. LoRA strength 1.0.