Sometimes 3 heads are better than 1? This LoRA usually outputs 3 heads when the weight is set to 1 and with ((three heads, 3head):1.4). Lowering the LoRA weight for img2img does help lower the influence on the model but you might need to inpaint necklines if you do. Sometimes you can get 2, 4, or more heads, which IMO is a bonus feature 😜.
Description
FAQ
Comments (10)
Blowjob, testicle sucking and anilingus at the same time
the heads that keep on giving...
Is it weird that I have a use for this one?
It ain't.
3 heads are better than 1
this so you can make the dude Sir Robin bravely ran away from?
Hey! Great job. I'm trying to train a similar Lora to reproduce Cerberus, the three-headed dog of Greek mythology. I have a good set of images to train this concept, but my final Lora is always a one-headed dog or multiple dogs in the same scene. I think it is a captioning problem. How do you write the captions for your model? Thanks
I think I made this one before I started paying more attention to tagging, so I didn't do anything other than run it through BLIP and just make sure the trigger word tag is first, ex: "three heads, 3head, a woman sitting on a bed." Another thing that could be important is to make sure the tags don't say "3 dogs" or anything that you wouldn't want to have to add to the prompt to make it work. What might help the most in your situation is upping the network dimensions and/or increasing the epochs. Higher network dimensions help get better results when models are resistant to your concept at the expense of being less flexible.
@_NSFW_Â Wow, thank you very much for your attention. I think adjusting the network dimensions and the number of epochs will do the trick.
@diegoluiz did you ever get this worked out? I've tried (and mostly failed) to get a couple models to produce a cerberus and a lora built for that would be amazing
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.