In the website collection last week, a little partner asked me if I had any AI painting tools?
Yes, of course. After all, AI painting has not been popular recently, but has been popular all year. However, if AI painting in the past few months only accumulates strength, and occasionally a small fire appears, then in recent months, AI painting has changed with each passing day.
However, although AI painting sounds very fresh, we have to face several difficulties. For example, most of the models are from abroad, and the Chinese description is not friendly. The simplest example is that 6pen Art has developed an AI painting that supports the Chinese description.
In fact, it uses the open-source Disco Diffusion model, but when running “eight sweaty BMW Mercedes Benz in the snow mountain”, whether it is “BMW” or “Mercedes Benz”, AI will obviously be wrong.
The red calipers are very smart, but no horse was seen. In essence, the description words were translated into English and then the paintings were generated. For the broad and profound Chinese, the AI “understanding ability” at this stage is not enough.
Not to mention the hardware requirements of AI painting, an application that consumes GPU computing power, so even if someone moves AI painting to the cloud, they will inevitably have to queue or exchange points for acceleration.
It is not smooth enough to play. Is there an AI painting tool that understands Chinese, has a large amount of money and can generate quickly?
Indeed, Baidu recently stepped down in person and released an AI painting model called “ViLG”. It supports Chinese without saying anything, and the free quota is also very large. A Baidu account has 500 painting quotas, which can convert words into pictures in less than one minute.
As for the effect? It’s a mule or a horse. Let’s pull it out for a walk.
No Chinese dispute
In ViLG, the specific operation is not so complicated. It is the same as the old three. Enter the description words; Choose painting style; Set the drawing size, and then go to the lower right corner to generate it immediately.
So we just need to polish the description. Let’s try the sentence “eight sweaty BMWs galloping in the snow mountain” first
Have you noticed that ViLG does not only generate a picture, but also generates 6 pictures at a time according to your keywords. BMW is a horse, and Mercedes Benz is an action. Although the details still need to be worked on in the description, it is obvious that the AI painting is to understand Chinese.
However, it seems that it is not good to use “BMW and Mercedes Benz” for the Chinese dispute. I decided to try the “Braised Lion’s Head” I just ate yesterday:
There is no wrong lion. This famous dish has been well restored, especially the first and fifth pictures. The effect is gratifying.
Of course, I also tried “wife cake” to see if AI can draw a wife for me. But from the result, without a wife, it’s all cake·
It seems that ViLG performs well in the Chinese dispute.
After trying the food, I want to test ViLG from two perspectives, one is the portrayal of the big scene, and the other is the specific person
在上周的网站合集下，有小伙伴问我有没有 AI 作画的工具？
有，当然有，毕竟 AI 作画可不是最近才火，而是今年一整年都很火，但如果说前几个月的 AI 作画只是蓄积力量，偶尔小火出圈，那在最近几个月里，AI 作画可谓日新月异，席卷而来。
但是，AI 作画虽然听着很新鲜，有几个坎，我们不得不面对，比如模型大多出自国外，中文描述不友好，最简单的例子，之前 6pen Art 搞了个支持中文描述的 AI 作画。
其实用的是开源的 Disco Diffusion 模型，但跑「八匹汗血宝马奔驰在雪山之中」的时候，无论是「宝马」还是「奔驰」，AI 明显会错了意。
红色卡钳倒是很灵性，马却没见到一匹，因为本质上，还是把描述词给翻译成了英文，然后再生成画作，对博大精深的中文而言，现阶段的 AI 「理解能力」不够。
更不要说 AI 作画这种极耗 GPU 算力的应用，硬件要求摆在那，所以哪怕有人把 AI 作画搬到了云端，难免还要排队或换取积分加速。
把玩起来不够丝滑，那有没有理解中文，额度又大，生成还快的 AI 作画工具呢？
还真有，百度最近亲自下场，出了一款名为「ViLG」的 AI 作画模型，支持中文不说，免费额度给的也很大，一个百度账号有 500 张作画额度，不到 1 分钟就能把文字转换成画面。
在 ViLG 里，具体的操作没有那么复杂，还是老三套，输入描述词；选择作画风格；设定画作尺寸，最后去右下角立即生成就 OK。
你发现了没，ViLG 不是只生成一张画，而是根据你的关键词，一口气生成 6 张，宝马是马，奔驰是动作，虽然细节上还需要在描述词里下功夫，但很明显，这个 AI 作画是看懂了中文。
当然，我还试了试「老婆饼」，想看看 AI 会不会帮我画出一个老婆来，但从结果上说，没有老婆，全是饼。·
试完了吃的，我想从两个角度测试 ViLG，一个是大场景的刻画，一个是具体点的人物、动物的描绘。前者是国外几个模型的拿手好戏，追求一个史诗感，后者在 3、4 个月前，还是 AI 作画的盲区。
大场景的刻画上，看来 ViLG 还是很给力的，再放两张竖长图，大家来欣赏欣赏，试试像素风，有惊喜：
The following is what I like to see. I didn’t think so much at first. I just measured the description of “midnight, girl, CG sense” in the cyberpunk version. To be honest, they are all Cthulhu who can’t be looked at directly. The only thing I can see is the following one: < / P > < p > < img SRC=“ https://www.toolpic.com/wp-content/uploads/2022/10/frc-f357fc743c00953f48f65eef9caaf990.jpeg “/>
Later, I found that the original addition of” delicate face “can make the little sister’s face normal and generate several reasonable pictures.
There is also this girl’s mecha with eight abdominal muscles:
< / P > < p > what really opened my mind was the little sister of anime animation, whose oil painting style of” long hair, high definition, delicate face, clear details, girl, anime, CG sense “: < / P > < p > < img SRC =” ” https://www.toolpic.com/wp-content/uploads/2022/10/frc-30717b74390cba8182e2e6a5046ce6a6.jpeg “/ > < / P > < p > realistic style” long hair, high definition, delicate face, clear details, little sister with beret, anime, CG sense “: < / P > < p > < img SRC =” ” https://www.toolpic.com/wp-content/uploads/2022/10/frc-8d3a3bd17c881b22df3ec956ace563d0.jpeg “/ > < / P > < p > anime style” high definition, delicate face, clear details, imperial sister, CG sense “: < / P > < p > < img SRC =” ” https://www.toolpic.com/wp-content/uploads/2022/10/frc-9937a4eb11141c910f493c6a3b846de3.jpeg “/>
Cartoon style” figure drawing, high-definition, delicate face, clear details, little sister with two ponytails, jk, high-definition cg sense “:
The little sister of national style in oil painting style is more unique:
If you want to play, please visit the link directly: