【画像が見えない人は、表示設定のX、XXXフィルターを外してみてください】
LORA作成代行のコミッションを受け付け開始しました。（値下げしました！）
詳細はプロフィールをご覧ください。
★ ★ ★ ★ ★ ★ ★
ぽっちゃり好きなので、Grokに適当なぽっちゃり露出高めのキャラを作ってもらいました。足先は作っていないので、お好みのタグを入れてください。
トリガーワードを使用せず、「○○_style」「official art, 」だけ使えばスタイルLORAとしても使えます。
もしもポーズが変わらないようで、「full body」タグが入っていたら外してみてください。
ADetailer使用推奨。
画像を作ったら、貼ってもらえるとプロンプトの勉強にもなりますし、励みになります!!
※商用利用・二次配布・無断転載厳禁※
----------------------
[If you can't see the image, try disabling the X and XXX filters in your display settings.]
I've started accepting commissions for creating LORA characters. (Price reduced!)
See my profile for details.
★ ★ ★ ★ ★ ★
Since I like chubby characters, I had Grok create a suitable chubby character with a slightly revealing outfit.
The feet are not included, so please insert your preferred tag.
You can also use this as a style LORA by using only "○○_style" and "official art," without using trigger words.
If the pose doesn't change, try removing the "full body" tag if it's present.
ADetailer is recommended.
If you create an image, please post it! It will help me learn about prompts and encourage me!!
Commercial use, redistribution, and unauthorized reproduction are strictly prohibited.
Description
FAQ
Comments (7)
A nice idea for a character lora. It seems to be somewhat overtrained though (I had to significantly lower the weight to get my style lora work along with it). Also consider pruning the trigger words in the model card, a lot of them are not necessary to correctly generated the character (especially facial expressions tags).
Oh... From my perspective, I think it's a good balance...😢
After consulting with the teacher who taught me the LORA learning method and with AI, I found that there were some flaws in the current learning method, so I will update LORA soon. I will also review the learning method going forward.
@lone_warrior_tiv You shouldn't feel to bothered about my comment. I just wanted to point out, where you maybe could improve. Also I don't really know much about training OCs. I assume you trained this with one image in the datsaet only? If that's the case, I guess it's difficult to figure out the correct training settings.
@RisingV @RisingV Gemini also drew the back view of this character. So, aside from the head, there are essentially only two source images.
@RisingV I've updated it; what do you think?
@lone_warrior_tiv I tried it and it adapts to other styles much better now. Good work! However one of the style loras I tried does not work good with chubby characters, it took the chubbyness mostly away. But that's the fault of my style lora.
I don't know how gemini is doing image generation (nano banana?). If it's possible, I would recommend to generate more than two images with different poses of the character for training. Then you can avoid training in the pose (but it works with different poses like it is now).
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.