FilthyDegenerate

Gwen - Total Drama Island [LoRA]

Added 2024-12-18 14:59:22 +0000 UTC

My god this bitch gave me TROUBLE, I'm a newbie but holy shit this was hard, so I noticed every jabroni on Civitai has a 213mb LoRA, well thats for a reason, higher ranks mean the AI will learn faster, specially 'tiny' details, namely her lips, which the upper is slighly darker than the lower, I was trying to save space making sub 100mb~ LoRA's but that costs accuracy, a lot of it, plus a high batch size seens to degrade the quality of the LoRA, after all we are getting more speed out of if, so is only fair to assume we are losing something in return, that being quality/accuracy.

Also CAPTIONING, manual specially, don't relly on WD1.4, do it manually, the difference in training accuracy is severely impacted if you do it manually, it takes longer? sure but you are more likely to get a good LoRA first try.

Also the zoomers at https://www.reddit.com/r/StableDiffusion/ oozing at HunyuanVideo when it does not support i2v yet, I'm impressed by the quality, but these VRAM requeriments are beyond stupid, we need less image degration when using i2v, we need the speed of LTX+STG (120 seconds to get a 4 sec vid on my 3090), and we need a good workflow AND we need an uncensored model than is more than 2B camera paning bullshit.

The i2v scene is evolving FAST, and NVIDIA knows that, this is the major reason they are gatekeeping VRAM, remember some months ago when I said here on my Patreon they would be VERY stupid to make a 28GB 5090? they didn't, but they didn't keep the 5080 on par with the 3090 either, AI i2v is about to change the porn scene and they know that so they want to keep the average joe (me and you) from having access to these giga vram GPU's.

https://civitai.com/user/BootyAttackArt