Realistic little people don't seem to be represented in the model. This Lora attempts to remedy that, at least for little women. The Lora was trained only on images of female little people, but it will probably create an accurate male little person as well. Works well with other Loras. Strength of 0.8 seems to be the sweet spot. You will probably get random tattoos on occasion. Ladies willing to bare it all seem to like tattoos. Your feedback and videos are appreciated.
Description
FAQ
Comments (5)
Only got bad result, can you share da workflow?
I just added the two workflows I use. All videos in the showcase were made with the modified Kijai T2V workflow. I also get good results with @bonetrousers 2step t2v. Were you using other Loras as well?
Maybe you're missing the keyword of l1ttl3?
@voyeur19 only this lora but I don't know what Kijai is I'm kinda new at comfy stuff I always used forge until recently =v="
@Kanonno88 Download the two PNG files I uploaded and drop them into your comfy. Kijia created the Hunyuan video wrapper. That workflow is an updated version of the example workflow from that release. The other workflow, 2step t2v uses the native comfy nodes. This lora should give good results with pretty much any workflow. I also included prompts with all of the showcase videos. Start with those as examples in your current workflow and modify from there. Don't forget to use the trigger, l1ttl3 and it works best at 0.8 strength. Higher strength doesn't result in as good motion. Note: Both of the workflows I'm using require Triton and Sage Attention if you use them as is.