Ahegao Face - Klein 9B/Z-Image
For the best results, use ahegao face in the prompt and pair it with descriptions like eyes crossed, tongue out, saliva dripping off tongue, to fully utilize the trained dataset.
Check my images to see varying settings that I used for both Klein 9B (Base and Distilled) and Z-Image/Z-Image-Turbo.
Description
FAQ
Comments (2)
What format is it? I’m getting errors when loading in diffusers.
Invalid LoRA checkpoint. Make sure all LoRA param names contain 'lora' substring.
Is it a simple key remap?
This was trained using OneTrainer, so the .safetensors file uses the standard Black Forest Labs key naming convention. ComfyUI reads this natively, but the diffusers library uses a different internal key structure for Klein models, which is why it is throwing that substring error. It is definitely just a key remap. You will either need to run a standard BFL-to-Diffusers key conversion script on the file, or if you update to the absolute latest version of the diffusers library, pipeline.load_lora_weights() is usually able to auto-map the standard Klein keys.