Recently, fans have sent me a lot of workflows they have done, asking me to help see "what went wrong". At the beginning, I was still very patient, especially when using the plugins I developed, and I felt obliged to explain these black box things to the users.
Gradually, I realized that something was wrong. Everyone's workflow has obvious commonalities (the mistakes are relatively consistent), and secondly, the code that is directly cv with a third-party coding tool cannot be run (especially code novices).
I thought about it for a long time and came to the conclusion that code novices are not suitable for development workflows
Ahaha, just kidding. The conclusion is thatNovices have no professional coding training, so they don't know how to develop it at all.Of course, development is not a martial arts secret, simply put, development isAnalyze requirements, disassemble steps, implement them step by step, and test and tuneprocess. The technical term for workflow development is called NoCode/LowCode low-code platform, which decomposes requirements into functional modules and then connects them through flows.
But beer is also wine, what do you say?
So, after thinking about it, should this matter be handled by Coze officials? As a result, guess what, the official is about nothing! In fact, it is easy to understand, the official provides a very detailed "manual", you can read the manual before developing. Or it is "copying", piling up a lot of cases, you copy it, copy it, and study it.
However, who started learning Chinese characters with the "Xinhua Dictionary"? Such a large pile of text looks very tiring. So, there is this article. I have to tell you the conventions of development (workflows). With these concepts, I believe that if you build again, it will be easierPass it all over the placeFinish.
Conventional
Let's start with the simple and easy to use, you can open your Coze Workflow and follow my step-by-step article.
1) Give your workflow a name that you can understand at a glance
The naming of the workflow is not arbitrary, when you want to create a workflow, you need to fill in the "workflow description" first, which is very critical, the description here is for the large model to determine when to call, so you can't fill in it blindly. What should I do if the description is not clear in a few words? Let Doubao help you, tell Doubao what you want, and tell Doubao, What kind of workflow do you want to create, help me generate a professional workflow description, and a streamlined workflow name, requiring that the workflow name only allow letters, numbers, and underscores, and start with letters." Of course, you can even ask him to help you generate a framework for your workflow, but that's not up here.
0) Once you have an idea, think about what stream you want to create, and then it's not too late to do it.
I deliberately arranged a step 0 here, which means to do it at the beginning, earlier than the name. Take a specific example. You can see how I analyze it.
I want to do a "Good Morning Daily" workflow and send good morning pictures to my WeChat fan group every day, as shown in the picture below
Let's analyze what needs to be done in this matter (the profession is called splitting SOP)
1. Trigger every morning to send messages to the WeChat group (send the generated images)
2. AIGC content analysis
1) Draw the background drawing with a sketchboard
2) The top left corner is the time and lunar calendar
3) In the upper right corner is the text "Good morning"
4) In the bottom right corner is the fission QR code
5) The middle area is a diagram
6) There is a "golden sentence" in the middle of the picture
So one question becomes three,"Dismantling and simplifying complex problems is the core ability of programmers”,
How do I automatically trigger workflow work? There are many ideas, such as linking the workflow in Feishu multi-dimensional tables with the current workflow, or directly telling the large model to execute regularly.
AIGC is simple, just stack the previous data through the drawing board into the final picture. But where does the content come from?
There is a very clever design, we asked the large model to speculate on the emotional color of today's golden sentence, and use the description of the picture to generate a SeeDream4.0 prompt, and then use this prompt to generate an image, directly doubling the efficiency!
So when our image generation is successful, where can I get this image? This involves a more interesting tool called uploader, when we have temporary files and images that need to be temporarily saved for subsequent large models and pipelines, then we have to create a URL link.
This is how "Golden Sentences of the Day" 1.0 came out, and it took me about 1 hour from the beginning of design to online release.
