Case description:
One day, a friend suddenly sent me a message and said: I see that your friend Quan has been showing off AI painting recently, which seems a bit awesome; my kid is participating in a national painting competition recently, can you let your AI help draw some interesting pictures? Creative painting. To inspire children, children can refer to the pictures drawn by AI to create more thoughtful words. "Then she sent me a picture (required by Tianshu)...
As far as I am concerned, the awesomeness has already been blown out, and I am still frequently rioting my friends in the circle of friends. Out of embarrassment and helplessness, I had no choice but to bite the bullet and take the job. The so-called bragging is really cool for a while, filling the crematorium. My bragging is so awesome that I have to fill it out on my knees...
After the background is explained, I get the message:
1. Ask me to help generate paintings with AI
2. Give a picture with a text description on it
Next, I will talk about my implementation ideas:
1. Do ocr translation of Tianshu into text information
2. Use AI for information extraction (abstract, keywords) of translated text information
3. Use the keywords that AI understands to generate spells suitable for AI painting
4. While generating spells, use AI-generated keywords to search for suitable painting styles
5. Start combining spells and styles into spells that will eventually let the AI paint
1. OCR decodes the scriptures to get the text
Here I used a few simple lines of code to complete it. In fact, you can use the result of dragging the picture without code to play. The specific web page:
https://huggingface.co/spaces/awacke1/Image-to-Multilingual-OCR
The world is so big and beautiful My home 8(0/##8[Mountain 4 states[「[「|0#00~1 {#|7|Call me[元{}5
"The Representative Office of the Food and Agriculture Organization of the United Nations in China participated in the hosting of "Beauty in Everything"
Erica Children's Painting International Tour Exhibition: United States: France: Italy: Russia: Japan and other countries are simultaneously soliciting collection time
February 208-April 208, 2023|Quanzhou Station Overseas Tour Exhibition|City to be determined Online tour exhibition will be launched simultaneously during the tour. Online 720 cloud real-scene exhibition submission mailbox
20_8<<0@126.< Drawing suggestion 1. Wildlife 2. Ocean. forest. The scope of collection of wetlands and other diverse ecosystems is open to children and individuals under the age of 12 from all over the world. Groups are allowed to submit contributions. Submission requirements 1. Painting size: 420 >2859 (big 8 format)
2. The submitted works will not be framed. No frame 3. The name and age of the work must be noted. countries and regions. Contact information
4.网上提交作品,入选后通知作者快递作品5。入选作品组委会拥有使用权评选标准绘画要求富有童真。
童趣围绕 "生物多样性保护" 主题,内容积极健康作品返还巡展结束后。
2024年4月前返还作品,颁发参展证书主办单位「40 @几4
(联合国粮农组织驻华代表处)中国晚报摄影学会中国儿童画国际巡展组委会支持单位中国关心下一代工作委员会儿童发展研究中心协办单位世界文化交流联盟启今集团~幸~0^公益支持业之峰装饰集团裒 堕 % /撑 少三 !关注"中德嘉美"公众号获取儿童画国际巡展相关信息守护生物多样
自然之美
2.把得到的文本信息作摘要和关键词抽取
这边我用了chatgpt来做；需要其实不用chatgpt写几行简单代码也行
如果考虑做AI应用产品的可以考虑代码来实现，链路更好串接。
3.把AI理解好的关键词用AI生成适合AI作画的咒语
这边我也是用chatgpt来直接把中文翻译成了英文，然后同步的让chatgpt帮忙生成了适合AI作画的描述方式
4.在生成咒语的同时，用AI生成的关键词检索合适的画的风格
到best prompt里面搜风格：
可以在百度文心搜索风格
5.开始把咒语和风格组合成最终让AI作画的咒语
不加关键词检索合适风格，直接用chatgpt生成的咒语作出来的画，如下所示：
加上关键词搜索合适风格后，做出来的画如下：
小结：
坑填上了，问题也给了个不算完美但是可以接受的解决。里面其实有很多细节是可以优化和精进的，比如画风选择，指定角色个数、色系修改。然而这篇文章重点不在于给大家介绍如何打造精细稳定的工业流水线，这些事交给天天996的人就好了。我只是想跟大家介绍，作为一般的人没有太多编程技巧，对计算机和各种奇淫技巧的算法了解也不多的一般人。可以如何去利用AI，可以如何把问题作简单拆解和流程拆解让AI为我们服务。
其实看下来也不复杂，Ai无非就是信息抽取能力、随机生成能力、非结构化的信息解析能力。我们要做的就是怎么把一个实际问题（更系统、更复杂）解耦开来，让AI为我们服务，人机协同。