Same thing as the PDXL one, but this is so it's showing on the SDXL generator.
Recipe:
1 cup sass
3 cups WHOOPS.
5 cups WHERES THE. CHEETOS?
Bake at 450 for 24 hours and then call the nearest fire department.
Read a few memes while you're in the ER for severe burns
(KIDDING.)
Contact Us:
Our Discord: https://discord.gg/HhBSvM9gBY
Earth & Dusk Media https://discord.gg/5t2kYxt7An
Backups: https://huggingface.co/EarthnDusk
Send a Pizza: https://ko-fi.com/duskfallcrew/
WE ARE PROUDLY SUPPORTED BY: https://yodayo.com/ / https://moescape.ai/
JOIN OUR DA GROUP: https://www.deviantart.com/diffusionai
JOIN OUR SUBREDDIT: https://www.reddit.com/r/earthndusk/
Description
Safetensors
FAQ
Comments (11)
Do you have a guide or resource for training embeds/TIs on Pony XL specifically?
Every time I've given it a go, it doesn't seem to actually train anything. In the extremely rare case it does, it seems like it reaches a point early on, and then stops. So all epochs produce the exact same result.
I'm familiar with Lora training, but this is really kicking my ass.
"EMBEDDING MERGE" is an extension i've listed in the description xD it lets you make things without training it.. I do NOT EVER Train my embeds - I make like 20+ of these in a night if i'm with extreme dumb ideas...
Which i have not really had in QUITE A WHILE
@duskfallcrew Ahhhhh, so these are mainly text prompts turned into embeds? Guess I was barking up the wrong tree.
My bad. There's so few people making embeds for Pony that I haven't had much success finding resources on training em.
Still, thanks for taking the time to answer, some of the other people I've asked just straight up ignore the question.
@anon4324 Because 90% of us are using embedding merge XD because training Embeddings on XL/Pony are not impossible but i haven't really bothered trying...
@duskfallcrew Yeah, I kinda figured that was the case, though I had hoped not and that someone else out there was training.
I just don't really understand how using merge and feeding it prompts is going to give a consistent result, or at least the result you expect.
@anon4324 LARGELY it depends. Aesthetic things like lighting, some color theory helps - but consistent characters are more just -- you NEED a lora for that sort of thing, but i've found key prompts and some scene ideas work
If you have a consistent character that a MODEL KNOWS an embed merge with that sort of theme would work, in the case of Sailor Moon 90% of the models know SOMETHING towards that, though i try nto to use it on realistic models because it will Do things i dont want it to
@duskfallcrew Yeah, that's kind of the thing embeddings are used for, and what I was trying to do. Take concepts the model knows, and just coax em out a bit better, or reinforce something. I've one success actually training an embed, and it visually does exactly what I want, but after that I haven't had any success, even with the same settings,.
Still, I've been messing around with EM though, and it's actually quite fun. I've been using it to combine commonly used tags into little chunks.
One thing I can't figure out, is how you'd add a negative prompt to the prompt EM uses. Like say you want to use a tag that has bleed with another tag, in normal prompting you'd just put the offender in the negative prompt and the bleeding is usually solved. I'm not sure how to do that with EM's prompt. It's not like I can just give the offending tag a negative weight, can I?
@anon4324 You just make a negative embed? XD
@duskfallcrew Can't really call a negative embed with wildcards. Well, you can call it, but it won't plop it into the negative prompt. Besides, having an embed that works without also needing a negative version to offset something just seems like good craft.
@anon4324 Eh i don't use a lot of negs, i do and i dont depending
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.