we're cooking with gas now bois

(what upsets me is this server costs 4x more to run per hour than I earned at my first job)

e/acc || Building @ /nova || Catgirl maxi @ /imgnai || Dogecoin core dev 2013/14

𝔗𝔞𝔨𝔢 𝔑𝔬𝔱𝔢𝔰 ✍️
𝔗𝔞𝔨𝔢 ℭ𝔥𝔞𝔫𝔠𝔢𝔰 🎰

𝖂𝖊𝖘𝖙 𝕳𝖆𝖒 𝕱𝖆𝖒𝖎𝖌𝖑𝖎𝖆


very much so. (that will make its money back if inference)

Prosumer is where inference is at my man - or consumer if you happen to have a 3090/4090 blower vendor, so I hear 👀 

This is training a flux lora, bout 41hrs left to go! (she's a big 'un, seeing how far we can go before a fine-tune is needed)

flux is so crisp, are you going for photorealistic portrait, or is it...animated?

imgnAI

Going for a sorta stylized look - think modern cg/anime sorta feel.

But pending I don't overtrain it (and pending LoRA-256 can handle it), it shouldn't lose the ability to do photorealism or anime via prompting.

I'm hoping it's a hole-in-one on training this but I'm basically doing a manual grid search for the ideal settings atm. Manually curated a few hundred images, manually tagged them to fit within T5 bounds, and there's 4 different config variations I wanna try it with.

So, thanks /imgnai for footing the bill on this one 💀