Type | |
Stats | 3,877 |
Reviews | (212) |
Published | Jul 22, 2023 |
Base Model | |
Hash | AutoV2 60D350220D |
欢迎大家使用本模型。如果你喜欢我的工作,欢迎在ko-fi上为我买杯咖啡☕
Everyone is welcome to use th models! If you like my work, feel free to buy me a coffee on ko-fi ☕.
如您有模型定制或其他合作、咨询需求,请联系Wechat: pomelonioi。请简单注明来意,时间宝贵不闲聊哈,谢谢理解!
If you have any model customization or other collaboration and consultation needs, please contact Wechat: pomelonioi . Kindly provide a brief description of your inquiry, as there's limited time and no time for casual chats. Thank you!
为了更好地让新司机上路,作为老司机的Muggle在鹅群(891417811)放了自制抽卡机(wildcards),包含日常穿搭、职业制服、搞怪戏服、节日礼服等数百套风格;
此外,如果你会Openpose,我还准备了300个用于控制人物姿势的Controlnet骨架图(站姿、坐姿、躺姿)方便又实用;主流插件也都整理好了,等你来取。
NSFW请加TG群 https://t.me/+bm9GZdmvf3dmMWM9
Join my TG group to chat https://t.me/+bm9GZdmvf3dmMWM9
【麻瓜lora】脸型大小调节神器
不知你对脸型是否有一些偏好?是更喜欢瓜子脸还是小圆脸?
一晚上跑图几百张,精选想要的几十张,然后一张一张PS修改脸型会不会很累?用修图软件修改是不是传来传去也很麻烦还尺寸有可能被压缩?
从专业角度来说,如今最流行的脸模lora使用方法adetailer插件只能替换人物五官,而它并无法置换底膜生成的人物脸型,这会大大减弱你修复脸部之后的相似性
所以麻瓜出品了这个脸型调节神器lora,合理权重范围是-1.5到1.5(往上取值也不太会崩,只是未必好看,根据个人需要调节)
负向权重的效果是:瓜子脸,肤色较白皙;
正向权重的效果是:小圆脸,肤色偏红润;
该lora不需要触发词,主要作用在脸型,对皮肤调节力度不大
经测试,在majicmixRealistic、墨幽人造人、moonfilm等写实模型上表现良好,不会对脸型和皮肤(一丢丢)以外有过多影响,叠加NWSJ、Koreandolllikeness等人脸lora的效果也很好;
在二次元模型如meinamix、anything上的作用力不强,如需调节二次元人物脸型可能需要加更高的权重,但麻瓜发现就算变化小也基本不会破坏画面构图。
该lora训练时长超过30个小时,制作过程实属不易,喜欢的话务必给我点赞、返图和评论,如有测试反馈欢迎评论区留言告诉我!后续会进一步进行优化!
Do you have any preferences when it comes to facial shapes? Do you prefer an almond-shaped face or a small round face?
Spending a whole night running hundreds of images, selecting just a few dozen you desire, and then individually editing each one in Photoshop to modify their facial shapes – doesn't that sound exhausting? Moreover, using editing software can be troublesome, with the risk of losing quality due to compression when passing files back and forth.
From a professional perspective, the most popular face model, Lora, when used with the adetailer plugin, can only replace facial features but cannot alter the facial shape generated by the base model. This could significantly reduce the similarity to the original face after modifications.
To address this, Muggle proudly presents the Lora Facial Shape Adjustment Tool. The reasonable weight range is -1.5 to 1.5 (going beyond might not cause crashes but may not yield desirable results, so adjust according to individual needs).
Negative weights will produce an almond-shaped face with fairer skin tones. Positive weights will result in a small round face with slightly rosy skin tones. Lora primarily focuses on facial shape adjustment, with minimal impact on skin tones.
Through testing, it performs excellently on realistic models like majicmixRealistic, moyourenzaoren, moonfilm, preserving facial shape and skin characteristics (to a minimal extent) while also working well when combined with NWSJ, Koreandolllikeness, and other facial Lora models.
For anime models like meinamix and anything, its impact is less pronounced. If you need to adjust the facial shape of anime characters, a higher weight might be necessary. Nonetheless, even with slight changes, Muggle found that it rarely disrupts the composition of the artwork.
The training process for Lora lasted over 30 hours, making its creation no easy task. If you like it, be sure to give it a thumbs-up, share your results, and leave comments. If you have any testing feedback, please feel free to share it in the comments. Further optimizations will be carried out in the future!
【麻瓜lora】笑容大小调节神器
在SD中传统控制笑容大小的办法是写入expressionless、smile、(happy,smile)、(grin,teeth)、laughing、(laughing,open mouth:1.3)等tag
不得不说切换表情的操作实在是繁琐至极,而且在laughing上的表现力经常很差,尤其是当使用人物lora时非常容易受到底模中laughing素材的影响导致面部相似度降低;
所以麻瓜出品了这个笑容大小调节器来代替繁杂的tag,另外部分人物lora在训练时可能缺少笑容素材无法做出很好的笑容表情,可以用这个lora试试看。
与上次发布的脸型大小调节神器不同,笑容大小调节器同时在写实模型和动漫模型上都有着良好的表现,并且几乎不会对原有的人物脸型、五官、皮肤有任何影响。
喜欢的话务必给我点赞、返图和评论,如有测试反馈欢迎评论区留言告诉我!后续会进一步进行优化!
In SD, the traditional way to control the size of smiles is by using tags such as "expressionless," "smile," "(happy, smile)," "(grin, teeth)," "laughing," "(laughing, open mouth: 1.3)," and others. Switching expressions this way can be extremely cumbersome, and the performance of "laughing" expressions is often lacking, especially when using the character "Lora," which is easily affected by the base model's "laughing" materials, resulting in decreased facial similarity.
To address this issue, Muggle has developed the "Smile Size Adjuster" to replace the complex use of tags. Additionally, for some characters like "Lora," who may lack sufficient smile-related materials during training, this tool can be used to achieve better smile expressions.
Unlike the previously released Face Shape Adjuster, the Smile Size Adjuster performs well on both realistic and anime-style models, with virtually no impact on the original character's face shape, facial features, or skin.
If you like it, please give it a thumbs up, share your images, and leave comments. If you have any test feedback, feel free to leave a message in the comments section. Further optimizations will be made in the future!