「ComfyUI」Flux.1：只需 4M，告别 Flux.1 “屁股下巴“脸，生成更加独特多样的面孔

不知道小伙伴们在使用 Flux.1 生成真实人物时，是否有关注到生成人物的面孔特征特别明显，尤其是下巴部位的特征尤为明显，不管我们提示词中输入欧美人还是亚洲人，下巴上的特征都没法去除。

我们这里戏称"屁股下巴"脸，感觉这描述还挺形象的。

这么明显的特征也使得我们生成的人物更容易一眼 AI，今天我们就来介绍一款只有 4M 的 Lora 模型，但是却可以帮我们有效的解决"屁股下巴"脸的问题。

好了，话不多说，我们直接开整。

最近有小伙伴问听雨如何使用安装，这里统一说一下，听雨这里使用的绘图工具是 ComfyUI，对 ComfyUI 不了解的小伙伴可以看公众号置顶的两篇文章。

好了，我们继续，文中用到的 Lora 模型会放在文末的网盘里。

我们先来看看不使用 Lora 生成的真实人物形象。

提示词：photograph An girl, facing the audience, close-up of her face, 50mm . cinematic, 4k epic detailed, 4k epic detailed, photograph shot on kodak, detailed cinematic, hbo dark moody, 35mm photo, grainy, vignette, vintage, Kodachrome, Lomography, stained, highly detailed, found footage

可以看到下巴上的特征还是很明显的。

提示词换成一个亚洲女孩：photograph An Asian girl, facing the audience, close-up of her face, 50mm . cinematic, 4k epic detailed, 4k epic detailed, photograph shot on kodak, detailed cinematic, hbo dark moody, 35mm photo, grainy, vignette, vintage, Kodachrome, Lomography, stained, highly detailed, found footage

可以看到下巴上的特征依旧明显。

然后我们加上这款 Lora 再来试试效果，提示词不变，可以看到"屁股下巴"明显没有了。

提示词：A striking, high-definition photograph captures the arresting beauty of an Asian woman with a refined, seductive countenance. The image is expertly framed to showcase her enchanting features head-on, emphasizing a close-up view of her flawless face and expressive eyes. Shot in cinematic style on 35mm film with a 50mm lens, the photograph boasts an epic level of detail that is reminiscent of HBO's dark, moody aesthetic. The image is richly detailed with texture from Kodak film stock and features subtle graininess, vignette, vintage staining, and Lomography-inspired elements to create a highly atmospheric feel. The overall effect evokes the timeless charm of classic found footage while also showcasing the woman's stunning beauty in exquisite detail, making it an epic portrait that captures both her allure and the evocative atmosphere of the photograph itself.

我们直接来进行对比吧，相同的提示，左边没有使用 Lora，右边使用了 Lora。

差别还是很明显的，提示词中指定了是亚洲人，明显是右边的图更符合我们亚洲人的标准。

提示词：In a captivating image that blends retrofuturism with science fiction and Victorian-era aesthetics, we find ourselves in the world of Steampunk. The subject is a vibrant and resourceful young Asian woman who gazes confidently out at the audience. Set against a simple yet evocative background, her close-up face serves as the focal point of the image. Her features are highlighted by intricate details: defined cheekbones, expressive eyes, and an air of determination that radiates from within. The woman wears distinctive clothing characteristic of Steampunk fashion—a corseted dress with exposed gears, a top hat adorned with brass accents, and boots with cogwheel patterns. Her steampunk ensemble is complemented by the intricate details surrounding her: steam-powered technology such as mechanical arms, airships, and machinery that blend old world charm with innovative design. The overall effect is one of a neo-Victorian society immersed in an industrial revolution powered by steam-driven machines—a captivating fusion of the past and future.

不使用 Lora 的情况下，大多数的出图都和左图一样下巴部位比较明显，总是感觉怪怪的，右图就自然很多。

好了，接下来我们来说说这个 Lora 具体怎么用，这个 Lora 不能直接权重 1 干上去，权重 1 反而特征更明显。

权重越高特征越明显，权重越低"屁股下巴"的特征越不明显。

以下 Lora 权重分别为：2、1、0

以下 Lora 权重分别为：-0.3、-0.6、-1

可以看到权重大于 0 时的，下巴上的特征都比较明显，小于 0 的时候特征逐渐减弱，但是也不是越小越好，可以看到权重为 -1 时，人物有些许变形。

这里推荐权重设置为 -0.6 左右效果比较好。

怎么样，Lora 虽小，但是效果很好。不过这个 Lora 模型只针对女性哦！

好了，今天的分享就到这里啦，感兴趣的小伙伴快去试试吧！

网盘链接：模型分享