Sign In

Rina's Robo Faces - Lora

Dec 4, 2023
Base Model
SD 1.5
Usage Tips
Clip Skip: 1
Strength: 0.5
Trigger Words

Don't use at 1.0 strength. start at 0.5 and adjust from there depending on how it reacts with your model of choice.

Edit: Also, I don't mean the name to be misleading. Like the previous models, if you want this to work really well on a face, it needs to be a closeup, inpaint, or ADetailed image. On a wider shot, it will robo-fy the rest of the body more than the face. Still fun for sci-fi images, but I don't want people to think that it's not working correctly. I just carried over the previous naming convention. The training images are almost all of faces and upper body, but it does its best on humans to make the whole thing a cyborg. Animals are harder (for me) to do, but I've gotten a few good results. I'll post more images with a wider shot as I have time later today, also including some other style loras (like the Avril Lavigne angel image [kVoidEnergy and Avril Lavigne - Lora] and the "leaning on a wall wearing a red cloak" image, which was made with a depth map from ControlNet RPG 5.0 Depth Map and the World of Brimstone Lora.)

Updated version of my older embeddings, trained as a lora with an updated an enlarged dataset. As I did with v2 of the embedding, older images in the dataset were either run through img2img, or txt2img via controlnet with the previous embedding (in this case, v2). In the time since I made the old embeddings, many new cyborg-type loras have been released, and they are excellent. There may not be a need for this type of thing from me anymore, but I use it regularly, and thought I would share, after some tweaking. This is actually v2 of the lora, v1 was not published. The original was too overtrained, and the dataset was a little corrupted because I had two images of cat-eared cyborgs in the dataset, and that feature popped up entirely too much in outputs it wasn't prompted in. You may still need to throw "cat ears" in the negative if you stumble across an otherwisse cool setting that includes that in the output, but I think I've tamed it down, as well as the egregious lens flare that crept in, too.

This Lora works well with add_detail type loras, but go easy on it, as there is a touch of detail enhancement in the lora already. an example would be

<lora:lorarrf2:0.5> lorarrf <lora:add_detail:0.4>

If you're trying to use a specific face in the output, include it in the base prompt at a low strength, and copy the prompt into the ADetailer positive prompt and edit the strengths. Put this lora down to about 0.2-0.3, and then adjust the chracter lora to regular strength, or a tad higher, as it will be getting blended into the base image.

I hope you enjoy it, I've had a lot of fun with it.