This is for all the Latino Waterdog enjoyers out there. Keep on pulling up in the after party, no one says. This was trained on CivitAI's trainer. Buzz would be appreciated so I can try and refine this more, with a better and varied dataset—that includes videos. Diffusion-pipe does not love me and it sure doesn't love my 8GB.More of a protoype test than anything. I do not wish to do anything extensive at the moment, so I thought to train through image-only and character first before I move to mixing with images and videos regarding specific motions. Expect varying results and may require a good number of trial and error to get something decent. It's a rough patch but a ""good enough"" one.<h3 id="what-other-thing-could-i-do-than-make-a-lora-that-focuses-on-capybara's-bysw972pb">What other thing could I do than make a lora that focuses on capybara's?</h3>While this was tried on Hunyuan 720p FP8, it was meant to be trained on Wan T2V 1.3B. There's a severe lack of T2V loras for us furries, and since I am currently testing the waters before I make a more serious, dedicated lora, I hope some of y'all would not mind generating some water dogs. I have to give love to my Hydrochoerus hydrochaeris somehow.This is in rough works and tested through 16 epochs and trained through Hunyuan's 720p FP8 model in CivitAI's trainer with 37 images and no videos attached. It is not perfect but it's alleviating. The feeling of creating your own video lora feels a lot more cathartic than image lora creating. I understand most furries are relying on Wan nowadays but I have heard some Hunyuan loras can work through Wan. Wouldn't hurt to try it out! This is merely the first version and newer (and hopefully better) versions will come to fruition.Originally, styles were included but did not respond well. Most likely because I tested the model on SeaArt with a random Hunyuan workflow, so let me know if you get different responses if you use this locally. It was planned to be, "In the style of [cartoony, semi-realism, realism]" but that doesn't really work when all your dataset images are 2D and 2.5D.<hr /> Feel free to begin your prompt with, "The anthropomorphic capybara...", "The video shows an anthropomorphic capybara..." or "An anthropomorphic capybara...". These are optional as the main trigger word is simply 'capybara'. It does not have to be in the front, middle, end, whatever.You can place the word 'capybara' anywhere. Be sure to emphasize body color, body type, gender, etc. Yes, it can do futa.Showcase examples are obviously cherry picked. Took several genes and parameter tweaking to get the results shown. If you want a basic outline to start with,[They/He/She] has a [voluptuous/skinny/fit/chubby/overweight] body type. [They/He/She] has a [color type] body with tufts of visible fur.

Files with SHA256: 7f21f3af218f0db1548f3a8a05a413381b4bef08544925c1c3cd347c13b24c30