Last year, a creator with zero painting foundation generated two sets of emoticons with the help of AI and successfully put them on the WeChat emoji store, which attracted widespread attention. After a year, AI drawing technology has once again made significant progress, and the generated emoticons not only take less time and have a better painting style, but also achieve higher clarity and simpler operation processes. Today, AI-generated emoticons have moved from “drafts that run through the process” to “master works”, and the threshold for creation has been further lowered, and almost everyone can become an emoticon artist.
In April last year, as a novice with no painting foundation at all, I generated two sets of emoticons with the help of AI and put them on the WeChat emoji store, and the tutorial was spread by thousands of people.
After a year, I created a set of emoticons again, and I was amazed by the progress of AI image generation capabilities, which can produce emoticons in seconds with almost no cumbersome steps, shorter time, better painting style, clearer accuracy, and simpler process. If AI emoticons a year ago could only be regarded as drafts that ran through the process, then today’s AI-generated emoticons can be called master-level works.
Then let’s write a 2025 version of the emoticon artist tutorial step by step, and feel what changes have been made in AI images in a year.
Review requirements for WeChat emoji stores
Officially put a set of emoticons on the WeChat emoji store and become creative works that users around the world can freely download and use, which requires the production of emoticons in full accordance with the review regulations of WeChat emoticons. This is a propositional composition.
The entrance to the WeChat Emoji Store user is in “WeChat App-Me-Emoticon”, where you can choose to add various emoticons to the chat window. The emoticon works we upload will also be presented here for users to download.
As an emoticon artist, the entrance to upload emoticons is on the “WeChat Emoji Open Platform”, the URL is https://sticker.weixin.qq.com, and all emoji works are submitted and managed here.
During the submission process, the WeChat emoji platform will make some review criteria for emoticon works, and I made a form to see at a glance.
During this year, there was also a change in the review requirements for WeChat expressions. A year ago, the number of expressions contained in a set of emoticons needed to be one of 8, 16, and 24, but now it is any number between 8~24, which also provides convenience for us to use AI to make emoticons, after all, AI draws, there is no way to control the number too accurately.
The changes are amazing
We see new advances in AI technology every day, but if you want to perceive it intuitively, it’s best to compare the effects before and after.
What does a product manager need to do?
In the process of a product from scratch, it is not easy to do a good job in the role of product manager, in addition to the well-known writing requirements, writing requirements, writing requirements, there are many things to do. The product manager is not what you think, but will only ask you for trouble, make a request:
View details >
If a year ago, AI generated emoticons, and there was a slight technical threshold, then now, it can be said responsibly that anyone can use AI to create their own emoticon works with excellent graphics.
Let’s compare some major changes in the meme production process.
1. The quality of raw drawings has been greatly improved
The most intuitive visual perception is the improvement of the quality of the work, and it is self-evident that we compare the previously generated AI emoticons with the emoticons generated this time.
For example, although OpenAI released its first multimodal model DALL-E, the main capability breakthrough is still in the deep inference model; Focusing on images, Midjourney and Stable Diffusion are more effective, but they are also very expensive to use. China’s DeepSeek R1, released in January 2025, is also a large language model without image generation capabilities.
Therefore, in the first half of 2024, although some image generation models have come out, the technology is generally immature, and it is still difficult to generate some high-quality images at low cost.
As can be seen from the comparison chart, although the shape of the emoticon works generated in 2024 is already decent, there are still a large number of action poses that do not conform to physics, as well as meaningless random lines, and the aesthetics of the image are relatively weak.
As the end of 2024 approaches, large image generation models usher in a wave of technological explosions. In May 2024, OpenAI launched the GPT-4o multimodal model, and in March 2025, it abandoned its dependence on independent DALL-E models and launched GPT-4o’s native image generation capabilities, detonating the global network through a series of Ghibli-style comics. Domestically, Dream AI released image model 2.1 on December 18, 2024, and Dream 3.0 in April 2025 and was deeply integrated with Doubao, and its drawing ability has been rapidly improved, which is basically comparable to GPT 4o.
We can see that in the works generated in 2025, whether it is color, style, shape, or details, they will have the effect of human designers.
2. The picture is clear, no need to enlarge it twice
The reason why it should be taken out separately is because in last year’s version of the AI emoticon tutorial, there is a key step to enlarge the picture through a special tool to improve clarity.
But today’s AI image generation tools can basically generate high-resolution images, such as the image resolution generated by Doubao, the longest side reaches 1328 pixels by default, that is, Dream supports up to 2K native resolution (2560×1440 pixels).
With such clarity, the generated images can basically be used directly and can cope with most usage scenarios.
3. No more pro prompts
In last year’s AI meme tutorial, the prompts used to generate memes were:
Prompt 1:
A cute little monster is drinking, Simple graphic features, thick and cute limbs, simple and cutemovements, Jean Julien. four cute poses and expressions, laughing, angry,speechless, sad. different emotions, multiple poss and expressions, illustrations,flatcolors, simple line 2d painting, popular artstations, digital art,pixel style cartoons, lineworks, 8k
Prompt 2
A cat drinks wine, 18 emoticons,multipleposes and expressions , different emotions ,different body movements and expressions, Style of Disney, Style of Pixar, 3D, C4D, candy color, neon palette style, realistic, clear background,Oc rendering, soft focus, soft light, high saturation, 8K, hdr, masterpiece, super detail, –ar 1:1 –s 300 –v 5.0
It can be seen that in the previous prompt, a large number of binding parameters and specific descriptions of the style were used, and at the same time, for better results, it also needed to be converted into English.
And the prompt words I used this year were:
Basically, pure natural language can command AI to complete the generation of high-quality images.
4. Text can be generated directly
For large models, image generation only needs to understand the picture, but once the text is needed in the picture, it needs to be very accurate, so in the past, the part involving text in the AI-generated image was often a mess of garbled code.
The GPT-4o native image in March 2025 achieves accurate rendering of text layout, but it only supports English, and Chinese layout still cannot be completed.
In April 2025, Dream 3.0 will achieve complex Chinese typography, artistic fonts (such as laser characters) and semantic control (such as “pencil handwriting”) through Chinese text generation technology, which is known by the industry as “a watershed in the field of Chinese AI drawing” and successfully achieved global leadership in the field of Chinese.
With this ability, the text part of the emoticon we generate no longer needs to be manually processed by design software, but can be directly generated in the emoji image, and can also achieve special effects such as word art.
Like this:
2025 version of AI emoji tutorial
In the context of technological progress, the threshold for creating a set of WeChat emojis is also rapidly lowering, and it can almost be said that “everyone is an emoticon artist”.
Now generate memes, only 3 steps.
1. Generate images
Use any tool of bean bag, dream, or ChatGPT to describe the emoticon effect you want to generate and generate images.
In the generation stage, the AI-generated image ratio is required to be 1:1, and the design style and image content can be freely defined.
During the generation process, there are two methods:
- First, generate an image, and then use this image as a reference image to generate multiple pictures with different actions (this method is recommended to use dream generation, although Doubao and Jimeng are one family, but Jimeng can process pictures more finely);
- Generate an image, further use natural language to “maintain the same image and style”, and continue to generate multiple images.
2. Remove watermarks and process dimensions
Since the images generated by Doubao and Jimeng will have platform watermarks, we need to remove the watermarks. There are many tools to remove watermarks, you can find many free tools by searching on the Internet, just choose a watermark removal tool based on AI technology. If you are a Xiaomi or Huawei mobile phone, it generally has the ability to remove debris watermarks.
After that, it is to process the picture to the appropriate size. We mentioned in the review standards of WeChat emoticons that the size of WeChat emoticons is 240*240 and the volume does not exceed 500KB. We can choose a tool that can crop pictures, such as Photoshop, Figma, Axure, Sketch, Instant Design, MasterGo, etc., or even use PPT, as long as the image is processed into 240*240, and then exported.
Then use the same method to make covers, banners and other sizes of materials, and do some simple stacking and cutting.
3. Upload WeChat emoji open platform
Next, it is to submit the WeChat emoji open platform for review.
WeChat emoji open platform website: https://sticker.weixin.qq.com
Upload the finished pictures one by one as required, and if you want to give a compliment, make two more pictures (appreciation guide image and appreciation thank you image), otherwise you don’t need to.
After clicking submit, you will enter the review process, generally speaking, 1-3 working days, there will be a review result.
After the review is completed, you can find your own emoticon in the WeChat emoji store on your mobile phone.
In fact, this time, re-experiencing the process of making emoticons by AI was originally intended to be a small surprise for partners, but in the process of production, I found that with the east wind of AI image generation technology, the small art genre of emoticons has been completely changed. Compared with a year ago, when the picture was repeatedly adjusted and changed for the title of “expression artist”, the production process this time was extremely smooth.
But I also found some real challenges. I saw the beauty of the AI pictures and was eager to try, but when I really wanted to start production, I was stunned, I didn’t have any good design ideas in my head, this feeling was like having a dragon slaying sword, but not knowing the knife skills, no matter how powerful the AI ability to generate pictures, if the controller had no ideas, AI could not play a real role.
In the era of AI, skills are no longer the limitation of creation, and creativity and inspiration have become the biggest barriers to good works.
Human creativity is priceless.
