LoRAs and TI embeddings

LoRA and TI embeddings loading

To load a LoRA, simply write it in the prompt as you would in Automatic1111. The sintax is the following lora:{loraname}:{weight}. Under the hood, plugin will detect this and generate a LoraLoader node for each lora found. The strength will be applied to both model and clip inputs, and the last LoRA loaded will become the new model input for both first and second pass KSamplers, as well as the CLIP for positive and negative prompt (LoRAS will be chained). If you want to alter this behaviour, you can always do so in the workflow tab.

However, there are some limitations:

• Lora name is case-sensitive, meaning that if my lora file name is Rcktraccoon.safetensors and I write lora:rcktraccoon:1 in the prompt, LoRA won't be found by Comfy.
• Lora files must be safetensors. Other extensions are not supported.

To load a TI embedding, follow the same steps as in ComfyUI.