Put "mona \(genshin impact\)" in the negative prompt. That is, if you're using Automatic1111. Characters are usually tagged with the ip in parenthesis in booru and the slashes are how you escape those characters without putting emphasis in a1111. (I edited this to use backslashes. I'm not quite sure but I think it's backslashes instead of forward slashes.)
Prompt edit her later [... :0.25] with other clothes described before. Like this SD will generate other clothes AT the start so it won't bleed later on due to prior conditionning (this prompt edit work on A1111 and for comfyUi idk)
the examples (on huggingface) weren't rly good but when i gave it a try myself (3.1) holy crap! this is amazing! almost everything I ever wanted in an anime model, wtff?
When I read that 'oldest' is '2005 to 2010', I truly felt my age XD Could one even try generating Hamtaro, or is he too old for this model? He was, indeed, a splendid rodent! ^_^
I tried the Animagine but couldn't get it to cooperate with me... I've had much better success with Blue Pencil XL and Artium (v1.0). Maybe it's my style of prompting, or wrong expectations, but I'm mostly gonna stick with Artium I think.
Amazing video, thanks for sharing! I only wish you talked about Vram usage as well. Most probably i can't use it with 6GB Vram but will still download it. It seemed like possible to disappear good lol..
@@NerdyRodent Some new features have been added such as face swap, in-painting, out-painting, a prompt tool (describe), as well as a guidance scale which can help solve grey/washed out color issues by increasing the value (normally an issue with non-VAE anime models) or overblown colors/glare/cartoony look by decreasing the value, and some other things have changed/been added not mentioned here. To vary things up, you could go over the anime preset for the Colab version if you decide to revisit it, and maybe the new advanced setting's usefulness.
Thanks a lot for the info. Can we merge this check point with sdxl turbo ? If so how can we go about it ? (The ratio if merging ) . If any one had tried , what was the quality drop like ?
@@fossil98 I see thanks for replying. So does that mean dream shaper sdxl turbo was trained on sdxl turbo rather than merging it ? If its different architecture how come normal sdxl lora work on sdxl turbo ?
Mona is a character from Genshin Impact. When prompted "Mona" Lisa the AI is using her clothes, anyway of negating that?
Try with the name of the game in the negative, or with the description of those specific clothes.
LMFAOO
Put "mona \(genshin impact\)" in the negative prompt. That is, if you're using Automatic1111. Characters are usually tagged with the ip in parenthesis in booru and the slashes are how you escape those characters without putting emphasis in a1111.
(I edited this to use backslashes. I'm not quite sure but I think it's backslashes instead of forward slashes.)
Just describe that clothing in the negatives
Prompt edit her later [... :0.25] with other clothes described before.
Like this SD will generate other clothes AT the start so it won't bleed later on due to prior conditionning (this prompt edit work on A1111 and for comfyUi idk)
7:06 is my favorite part of all nerdy rodent videos combined.
Didnt expect to hear a voice crack
I kept replaying 7:04 for 10min straight, no joke, its brilliant.
😉
came here to learn about an anime model, ended up learning an entire model evaluation pipeline. this is why i sub, thank you!
Glad I could help!
Cool find. This one is definitely worth featuring.
That “newer, best quality” Mona was turning into the Genshin character “Mona”
I am VERY much looking forward to this, thanks for the work you've been doing!
Oh wow! This is amazing. Thanks for the prompting tips too, I think you're right about the negatives
Glad it was helpful!
wow this model is amazing
the examples (on huggingface) weren't rly good but when i gave it a try myself (3.1) holy crap! this is amazing! almost everything I ever wanted in an anime model, wtff?
what about 60s/70s/80s/90s art-style?
When I read that 'oldest' is '2005 to 2010', I truly felt my age XD
Could one even try generating Hamtaro, or is he too old for this model? He was, indeed, a splendid rodent! ^_^
Several decades older than that...
Several decades older than that...
1cow, caused a chuckle.
interesting that it sampled from Genshin Impacts 'Mona' character for it's mona lisa rendition
My god! It’s full of anime 😉
Thank you so much for recommending!! 🤩
Of course!!
negative prompts are reducing latent space and image variety, they should used with care
It's funny how many tokens are wasted on "quality" tags other mysticisms xD
@@ImmacHn Hmm... Never really thought about it, because no one ever Talks about it, even though it is pretty obvious.
I tried the Animagine but couldn't get it to cooperate with me... I've had much better success with Blue Pencil XL and Artium (v1.0).
Maybe it's my style of prompting, or wrong expectations, but I'm mostly gonna stick with Artium I think.
Did you try it with the Fooocus tool?
Is there a stable diffusion model that creates non-anime comic book style renderings, for those of us plebes that aren't into anime?
Hehe great video you made me laugh on your reaction on the cow the little silence before the big ohhh hahaha hilarious thanks again for the video!
Amazing video, thanks for sharing! I only wish you talked about Vram usage as well. Most probably i can't use it with 6GB Vram but will still download it. It seemed like possible to disappear good lol..
More VRAM is best, but SDXL is known to work on 4GB VRAM and up!
@@NerdyRodent Thank you, i will try it now 👍
Good model, but difficult to master, beginners are better off starting with something like a blue pencil. Thanks for the clarification rodent🐭👍
That Mona Lisa, is just Mona and Lisa from Genshin impact.
Are you considering to make a video of the new A1111 1.7?
You should go over the fooocus AI tool.
I guess I could go over it again, what do you think in particular - especially compared to my other videos on it?
@@NerdyRodent Some new features have been added such as face swap, in-painting, out-painting, a prompt tool (describe), as well as a guidance scale which can help solve grey/washed out color issues by increasing the value (normally an issue with non-VAE anime models) or overblown colors/glare/cartoony look by decreasing the value, and some other things have changed/been added not mentioned here.
To vary things up, you could go over the anime preset for the Colab version if you decide to revisit it, and maybe the new advanced setting's usefulness.
3.1 is just out!
Woohoo!
Bro I have an andriod phone, any solution 😢
Thanks a lot for the info.
Can we merge this check point with sdxl turbo ?
If so how can we go about it ? (The ratio if merging ) .
If any one had tried , what was the quality drop like ?
Not possible, different architectures of model.
@@fossil98 I see thanks for replying. So does that mean dream shaper sdxl turbo was trained on sdxl turbo rather than merging it ?
If its different architecture how come normal sdxl lora work on sdxl turbo ?
Could we have a few capybaras please?
Good rodent choice!
*munches on one of the carrots*
so its a simple another model for SD A1111 is just that ???
Nah not as good as NAI v3
It's closer than anything else though. Would kill for nai v3 to get leaked.
Hehe great video you made me laugh on your reaction on the cow the little silence before the big ohhh hahaha hilarious thanks again for the video!