This is a pony lora for the character Cooler. Cooler is the main villain from the Dragonball Z anime movie Cooler's Revenge, and the brother to Frieza. A weight of 0.8 to 1.15 works well with this, with most examples working best at 1. Sampling Steps of 20 to 30 works well with most models, as does a CFG Scale of 6 to 8.
Prompts for this model are Cooler, Dragonball Z, muscular, muscles, male focus, purple skin, colored skin, red eyes,no humans, tail, and face mask.
In some cases the models likes to make the chest armor purple or blue in some parts, so putting down white chest armor, or white chest bio armor, will help with that. Also, due to a few pictures used was during a part in the movie with floating rocks, as well as Stable Diffusion liking to think Dragonball Z means floating balls, I would put down prompts like floating rocks, floating balls, and similar prompts in the negative prompts section.
In a surprise find, removing or not having the prompt Dragonball Z used in the positive prompt area will result in the image becoming a character based on Cooler but otherwise something completely different and edgy in a cool way. Rather good for perhaps scary aliens, monsters, robots, and other characters with sinister looks and science fiction armor.
Description
FAQ
Comments (21)
Awesome, love cooler.
Thank you. I do as well, and Cooler is one of my first loras that I made last year. My second one in fact.
@adondlin255 Well you did a good job, this one looks pretty accurate
Honestly, if you still have the dataset, I would re-train it and make sure every image that is pretty accurately Cooler is tagged as him, and removing the dragonball z tag entirely. That's two tokens anyways, so it won't really be doing much; but you can't remove it because it seems to be tagged across like everything.
I suggest keeping your tags with pony minimal, basically start with the character name for most images, and then just a few for important details you think the ai may have trouble with or to separate differences in character between images. Like the white chest armor, purple body, red eyes, 1tail.
Right now for example, that dragonball z tag is holding like 90% of the character accuracy on it's own, may as well have the name do that. To avoid pony thinking you are talking about like, a physical cooler as well you could go like "dragonballz_cooler" and it would be one token.
Anyways, please keep at it, I am curious to see what else you end up making for pony ^^
@pihlawrkr738 I will try that. I also used One Trainer to make the lora and chose the option to use one file with prompts that applies to all images. I will remove the one prompt, and perhaps even tone it down more.
@pihlawrkr738 Also a good idea with the tag idea. I was thinking at first of using C00ler instead of Cooler, yet I went with the Cooler one since most people would write it out as such. Adding the dragonballz_cooler idea is good and I will try that. I am just a bit irked that the One Trainer doesn't add the metadata and I must write the prompts out and hope people read the description. I am also not sure if I am simply missing an option that affects the metadata.
@pihlawrkr738 I have made a new lora while trying the suggested prompt DragonballZ_Cooler. The results is similar it not near exact to the effect if the term Dragonball Z is not used as a prompt. At best, it combines the effects of when it is used and not used, blending the looks to a degree. All in all, it didn't work out as hoped and intended. As odd as it is, one must live with the odd situation of the picture looking radically different from the use or non use of the Dragonball Z prompt. A pitty really.
that is amazing Lora, If possible would you provide the dataset as well
I do not mind. I am just wondering how one goes about sharing the dataset and share it on this page.
Also, thank you.
Im happy to see some good Cooler love! I made a lora of him too! I remember using your original cooler lora back in the old days before Pony existed lol
Thank you. I see you made a number of nifty loras yourself. I already downloaded most of them already. I was thinking of making an SDXL version, but SDXL generally does not make decent character loras. The one I made for Belphemon didn't turn out as good as I wanted. Soon I would be caught up with all the new loras to match the old ones. I saved Belphemon for last.
@adondlin255 What do you train your loras on? I train all of mine on civitai since my computer is not capable of running stable diff lol
@SuperBubble6 My old computer was getting to be 9 years old, so I took the timely moment to get a nice computer with a 4090 graphics card. I got it on a Black Friday sale and left out as much as I can to bring the price down. Even so it cost a ton, yet it is an investment that should last a long while. I use that, along with the default SDXL (in its latest form) along with the default Pony checkpoints. For the trainer I use One Trainer for it is easy to set up and use, once one figures out what to do with the settings, names, and file locations.
@adondlin255 that actually sounds great tbh! My graphics card is over 10 yrs old and I have only had a job for a few months now. Its gonna be a while before I get a new card. maybe i can wait till cyber monday
@SuperBubble6 When you do, I advise to get an i7 instead of an i9. I found out after the fact that there is scarcely any difference between the two chips to the point of not being worth the difference in costs. Also, get one with a simple hard drive to keep costs down. Internal hard drives can be added later, and external ones tend to be cheaper. Lastly, get the regular HDD since a SSD kind, while faster, has an expiration date as the insides will bloat over time like a used battery.
Amazing! Your Dragon Ball LoRAs are fantastic! If possible, could you share the training images for each Dragon Ball LoRA you’ve created?
I think I shared the ones I did recently. I am not sure where to find where I put the images of the ones I used for the older models.
@adondlin255 Yes, that is why I am asking you to share this one's dataset.
@krigeta The moment I get my main computer up I will try to remember looking for the data set. For now I have been using my laptop till my family can get a new home.
@adondlin255 oh np then and wish you luck man! god bless you fam
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.