原文链接:【TheMisto.AI】Flux最强线稿模型实际效果测评(附安装方法) (chinaz.com)
不知道有没有小伙伴去测试一下哈,上一篇文章用的都是官方提供的参考图,经常关注Flux的小伙伴也知道那些ControlNet买家秀和卖家秀基本是两回事。所以这次我们测试到底一下好不好用,有没有效果。
安装过程
由于节点还没有被ComfyUI的管理器收入,所以我是手动安装的
项目地址:https://github.com/TheMistoAI/MistoControlNet-Flux-dev
下载我们的压缩包解压到ComfyUI\custom_nodes文件夹内,重启ComfyUI即可
模型下载
模型下载:https://huggingface.co/TheMistoAI/MistoLine_Flux.dev/tree/main
下载完放在下图文件夹内,这个TheMisto_model文件夹需要自己创建一个
工作流搭建
其实这个工作流是不用我们搭建的,官方提供了一个示例的工作流,就在刚刚下载插件的workflows文件夹内,我们加载一下就可以
然后这个工作流,它是用了个上次字节开源的16步加速的Lora,我没下这个Lora所以被我禁用了,当然你也可以去下一个测试一下。
效果测试
官方说这个模型对提示词要求很高,必须好好写,所以我这里用GPT帮我写了一下
我先测试一个小车的效果,这个Controlnet配合Flux还是挺占显存的,而且生成一张图的时间也挺久的,不过这看着效果还行
提示词:In a bustling cyberpunk city, a sleek, futuristic car with neon blue underglow cruises down a rain-slicked street. The vehicle's chrome surface reflects the holographic billboards and vibrant city lights. The street is lined with towering skyscrapers adorned with glowing advertisements in multiple languages. The car's design is angular, with sharp edges, and it's equipped with illuminated wheel rims and a pulsating energy core visible through its transparent hood. Dense fog swirls around the ground, and the headlights cut through the mist as the car navigates through the crowded, dystopian urban landscape. Pedestrians with augmented reality visors and cybernetic enhancements walk along the sidewalks, while drones hover overhead, monitoring the scene.
再测试一个美女图片,提示词我用美漫风格:A stunning female character portrait in classic American comic book style. She has sharp, defined features, with high cheekbones and full lips, highlighted by bold, black ink lines. Her wavy hair cascades over her shoulders, with rich, contrasting colors typical of comic art—deep reds or vibrant blondes. Her eyes are large, expressive, and framed by long, thick lashes, capturing a confident, yet approachable demeanor. She wears a form-fitting outfit with a subtle texture, reminiscent of superhero attire, and a slight smirk on her face adds a touch of mystery. The background is simple, featuring dynamic, diagonal lines or a gradient, enhancing the focus on her powerful, captivating presence.
还行是还行,不知道为啥画面有点横线
再试一张耳机:
A sleek, modern pair of over-ear headphones displayed against a futuristic, sci-fi backdrop. The headphones feature a metallic finish with smooth curves and soft, cushioned ear pads that exude comfort and high quality. The background is a digital landscape with neon grids, floating holographic interfaces, and a soft blue and purple glow. A faint, abstract representation of sound waves and futuristic symbols surround the headphones, highlighting their advanced technology. The lighting is dynamic, casting subtle reflections on the headphones, making them the focal point amidst the high-tech ambiance. The scene conveys a sense of innovation and premium audio experience.
效果也是不错的,都有遵循轮廓
继续测试一张室内效果:这个效果就比较一般了
A stylish, contemporary living room interior with a focus on clean lines, minimalistic decor, and a neutral color palette. The space features a large, plush sofa with sleek metal legs, positioned against a white or light grey wall adorned with abstract art. A low-profile coffee table with a glass top and metal frame sits on a textured area rug, adding subtle warmth to the room. Floor-to-ceiling windows flood the space with natural light, offering a view of a manicured urban landscape. The design incorporates modern lighting fixtures, such as a geometric chandelier or recessed lighting, casting a soft, ambient glow. Accents of greenery from potted plants bring a touch of nature indoors, while built-in shelves display curated books and art objects, enhancing the room's sophisticated yet inviting atmosphere.
我换了一个提示词再试试,我用JoyCaption反推了原图的提示词再次生成,效果好了不少,所以提示词还是非常重要的
当然也有失败的情况
我这边测试了个比较抽象的图片+joycaption反推,就挺奇怪的
总结
能用,但是离SDXL那么好用还有点距离,大部分效果还可以,没有特殊需求不着急使用ControlNet的小伙伴可以再继续观望观望
我们的网站有丰富的免费教程:
AI教程_深度学习入门指南 - 站长素材 (chinaz.com)