This model was created to provide a more full real looking kiss.

It was trained on 104 realistic and royalty free images of people kissing - usually just two people - at 5 epochs for 520 steps locally, using OneTrainer and captioning with Florence2.

Trigger Words: kiss, kissing, kisses.

While it can produce non-human and non-realistic images of kissing, the model tends to handle regular people or humanoids better.

Creator tested the model at many different steps, resolutions, and schedules but I do not see any settings necessarily being better than others when using this.

Recommended following settings but would encourage you to explore beyond these:

Model: Flux1-dev-Q4_0.gguf

Scheduler: Euler

Steps: 20

Guidance: 3.5

Resolution: 1024x1024 or 728 x 1024

Within each example image is also an example workflow for ComfyUI, just drag and drop the image into your ComfyUI, select the ÆMS Kissing LoRa in the LoRa Loader node, and your all set to go for testing.

https://civitai.com/models/881864/flux-two-people-kissing-lora?modelVersionId=987191