Trained on 79 images drawn by SeeDark (Jeong Hoon Park), the art director and original artist of PangYa. He passed away from leukemia in 2009. He was 29 years old. This was made out of love and respect for his art. I can't read/speak/write Korean, sorry. Please use an online translator.
パンヤの元公式絵のアートディレクターとお絵かきさん「SeeDark」様の絵79枚を参考としてこれを作りました。シードアーク様は2009年で白血病からいなくなりました。彼はたった29歳でした。彼の作品にの愛と尊敬としてこれを作りました。
Pony v3: Finally got a decent result, using this tutorial. I guess it was either because I wasn't using Lycoris, or because I used the score/source tags. Note that the style isn't always obvious, eg. the Karin/Ganyu/Mona pics in the gallery. Maybe because the dataset isn't very big?
Pony v3: やっと良い結果が作れました。このチュートリアル使いました。問題はLycorisじゃなかったからか、scoreとsourceタグを使ったからか、よくわかりません。今でも、なぜか絵のスタイルがはっきり表現されない場合があります。例:ギャラリーのモナ、カリンと甘雨の画像。データセットが割と小さいからかな?
SD1.5 v11: Trained with NAI. v1.0 was with AnyLoRA. I used newer methods and settings for v11, so it should give better results. I don't know why it's a Lycoris. I just used the same settings I used to make LoRAs.
SD1.5 v11: 今回はノベルAIでトレーニングしました。1.0の時はAnyLoRAでした。v11には新しい方法と設定で作ったから、前より良い結果が出るはずです。なんでリコリスになったかわかりません。いつものLoRA設定を使っただけですから。
Recommended weight(おすすめウエイト):
Pony v3: 0.8-1.2
SD1.5: Results vary, but 0.5-1.0 seem to be good. This mostly affects characters. I haven't tested with other models or male characters. May be difficult to use with other LoRAs, but you may be able to get around it by using detailed prompts, IMG2IMG, and/or ControlNet.
結果がいろいろ変わると思うけど、0.5-1.0は良さそうです。キャラぐらいだけ影響されます。他のモデルや男キャラには試したことないんです。他のLoRAと同時に使うのが難しいかもしれないが、詳細を説明するタグやIMG2IMGやコントロールネットを使ったら行けるかもです。
v11 Training settings used(v11のトレーニング設定):
Model: NAI
Batch Size: 2
Epoch: 20
Max Train steps: 2500
Save every 5 Epochs
Mixed/Save precision: bf16
CPU threads per core: 2
Seed: 123456789
Cache Latents
LR Scheduler: Cosine
Optimizer: DAdaptation
Max grad norm: 1
Optimizer extra arguments: "decouple=True" "weight_decay=0.4" "betas=0.9,0.99"
Learning rate: 1
LR warmup (% of steps): 0.1
LR number of cycles: 3
Max res: 768,768
Enable buckets
Minimum bucket res: 256
Max bucket res: 2048
Text Encoder learning rate: 0
Unet learning rate: 0
Network Rank (Dimension): 32
Network Alpha: 16
Scale weight normies: 1
Network dropout: 0.3
Rank dropout: 0.2
Module dropout: 0.2
Keep n tokens: 0
Clip skip: 2
Max token length: 75
Shuffle caption
Memory efficient attention
Crossattention: xformers
Min SNR Gamma: 5
Don't upscale bucket resolutoin
Bucket res steps: 64
Noise offset type: Original
Noise offset: 0.05
Adaptive noise scale: 0.005
The other options are probably default.
v1.0 Training settings used(v1.0のトレーニング設定):
Model: AnyLoRA
79 images
5 Repeats
Batch 2
Epoch 20
bf16
Seed: 31337
Learning rate: 0.0005
Cosine with restarts
5% warmup
Text Encoder learning rate: 1e-4
Unet learning rate: 5e-4
Network Rank (Dimension): 8
Network Alpha: 4
Convolution Rank (Dimension): 4
Convolution Alpha: 1
Max res: 576, 576
Min SNR gamma: 5
All other settings should be default.
Since the moment I first saw LoRAs, I've always wanted to try making this. It's my first style LoRA. It took 3 days of trial and error. LoRA/LoCon failed many times: grotesque body parts, image changing too much, background becoming ugly, etc. I was about to give up and tried LoHa once. It was a success.
初めてLoRAを見た時、これを作ってみたかったんです。初めてのスタイルLoRAです。3日間ずっと実験してしました。LoRA/LoConなら何回失敗:グロい身体の部分、イメージ変わりすぎた、背景がゴミになった、とか。諦めるところでLoHaを一回試してみたら、成功でした。
I used images from Danbooru, Google, and also some I saved when I played PangYa 20 years ago. I also used IMG2IMG to resize and clean some images. I did not include images that were too different from his Pangya art style.
借りたイメージはDanbooru, グーグル、そして20年前パンヤをやった時に保存したイメージを使いました。IMG2IMGでイメージをリサイズと綺麗にもしました。パンヤの絵よりスタイル違いすぎるイメージを使いませんでした。
Description
FAQ
Comments (18)
Thanks for beloved childhood memories. He passed to early....
I used to love playing this game, till it shut down >_< but thanks for the memories, and the history lesson, didnt know he passed away.
Try the recently released Pangya Reborn, a private server based on the latest S8! There's a small but dedicated community there.
Based! We need a dedicated Kooh lora now.
I wish this game would come back (Pangya M, where are you..) .. playin Pangya Celebrity with a tiny community doesnt hit the same..
I'm playing on Pangya Reborn... is the Celebrity server still around?
Sorry, I meant Pangya Debug. Pangya Celebrity I think is gone. Is Pangya Reborn better than Debug ? Supports English?
@datSato Yes, Reborn is in english and doesn't have the strict limitations Debug has, e.g. the automatic cancellation of accounts after a period of inactivity. Seems also to be more stable, and the staff on Discord is friendly and helpful.
@varthall01890 How is the activity of the player base? On Debug the server owner is regularly doing tee time with GM events and other things to get people online, limited rewards etc. I'd be willing to switch if Reborn is actually pretty active with players maybe..
well, and if they give you enough free currency to buy outfits and stuff like Debug does. I don't want to grind all over again unless its on a new game with active players lol
@datSato On Reborn the new accounts receive 4K pangs and 300 cookies, so plenty of currency for items etc. There are roughly 20 players during every GM event, which are held ca. every 5 days: https://pangya.community/t/gm-events-on-pangya-reborn/257 How many players are on Debug?
@varthall01890 ah, Pangya Debug basically gives you 500mil pang and 100mil cookies/points (or more), you can buy literally everything you want other than the items that are crafting only or etc. You get a ton of tokens for the gacha as well, with the option to donate a few dollars to the server to get more, or exclusive/rare cosmetics and etc.
I see during the GM events on Discord, they have enough players that they need to do multiple lobbies to get them all. It seems like there are more players on Debug. I'm sure both games are quite dead outside of the GM events, with just a few loose players online.
Hi, thanks for creating this, I was thinking to do the same but I'm still a beginner in generating AI art. Looking forward to see more art like this, it will be especially useful for private server's maintainers for banners, launcher images etc., of course with the permission of their authors. I have linked your page on the related Pangya AI art thread on Pangya Community: https://pangya.community/t/messing-with-ai/263/5
I'm nowadays playing on Pangya Reborn, there's a small but really dedicated community. Is Celebrity's server still up? AFAIK they were using a S4 server, while Reborn is based on the latest S8.
Playing PangYa Debug now. Celebrity is dead I think. Does Reborn support english?
@datSato Yes, it's in english: https://www.pangyareborn.com/
Hi. Glad you like it. Feel free to use it for whatever. I set up a server for myself on my own PC, so I don't know about Reborn/Celebrity. If you want to try setting up your own, there are server/client files and lots of tools floating around the net. Shouldn't be hard to find, and it's not hard to mod, either. Just watch out for viruses.
@Obliqeray Thanks, unfortunately I don't have a powerful PC, so I rely on external services. I have uploaded your model to seaart.ai, I'm waiting it to be approved. I might one day setup my own server to see how it works, otherwise I prefer to play on public servers so that I can play with other people. On Reborn there are roughly 20 players during every GM event.
BTW does your model include the names of each character, so that they can be used in the prompt?
@varthall01890 No, it's strictly style only. You'd need to either describe the character or make LoRAs for them. Edit: Actually, I remember some of the images auto-captioned with "Kooh". So some character prompts may work.
RIP to the great













