In this wave of AI boom, we are exposed to more B-side applications, such as waste processing, medical care, and even the breeding industry, and entrepreneurs have improved their efficiency. However, the implementation of C-side applications has always been limited to a few directions. Recently, a16z, a well-known American technology venture capital company, ranked the 50 AI products with the highest traffic on the market that are open to C-end users. With reference to this ranking, this article just makes an observation from the above perspective.
However, it is worth noting that although they are all open to the C-side, the main users of some of these products may still be designers, marketers, and individual developers in the enterprise, such individuals with professional status.
a16z’s ranking of the top 50 AI product traffic open to C-side | Note 1: The ranking data mainly refers to SimilarWeb’s website traffic data as of June 2023. For products with mobile apps, a16z also estimates (the method is MAU multiplied by the average number of monthly visits per user) and counts the traffic data together with the website data.
Note 2: The list does not include products without web pages, nor does it count the traffic of third-party platforms such as Discord servers.
There will be some simple conclusions in this part:
1. One product, ChatGPT, accounts for 60% of the top 50 traffic;
2. In the AI track, even the leading ChatGPT and the leading products of the "classical Internet" cannot compare in terms of traffic;
3. Apart from ChatGPT and its imitators, the fastest growing traffic is AI companions and AI content tools (these are also the two most common areas for overseas entrepreneurs).
If readers are willing to see more detailed data, they can continue reading, or they can jump directly to our product introduction chapter.
Traffic of each product in the Top50｜Image source: a16z
Overall, although it is on the Top50 list, the first-ranked ChatGPT accounted for 60% of the traffic, and then the traffic dropped off a cliff. The second-ranked character.ai's traffic was only 21% of ChatGPT, and the traffic of other products was even less.
But in fact, overall, the traffic of the Top50 is not low. Lexica, ranked 25th, had 560W visits in August, while DeepSwap, ranked 50th, had 696.4W visits in August, surpassing the 25th place.
Traffic comparison between ChatGPT and other mainstream products｜Image source: a16z
Comparing ChatGPT, which has the highest traffic among the Top 50, with other current mainstream products, ChatGPT is probably on the same level as Reddit, LinkedIn and Twitch, and far lower than WhatsApp, YouTube, Facebook and other products. It can be found thatThe AI track is still in its infancy, there is still a certain gap between head products and mainstream Internet products.
Traffic proportion data of AI products and growth of other categories except AI assistant tools｜Image source: a16z
Back to the AI track itself, from a category perspective,ChatGPT-like AI assistant categories accounted for 68.7% of traffic (including ChatGPT).In addition to ChatGPT, representative products include Google’s Bard and Quora’s Poe, both of which are in the Top 5 on the list.
But excluding the AI assistant category, in the past year,The number of user visits to AI generation tools and AI companions is growing rapidly, and they are considered potential categories.
Chart of the proportion of Top 50 consumer AI products classified by function. Note: Runway, ranked 32nd on the list, has both AI picture and AI video related functions. However, because its main model is Gen-2 as a video model and it has launched an app for the AI video generation function, it is classified as a video tool here. Functions related to image generation and editing will be introduced in the fourth part.
Because a16z’s category classification is relatively coarse-grained, in order to observe it more clearly, the author divided the Top50 products according to their functions.
We found that in terms of function, AI tools for video, audio, pictures, and text accounted for the highest proportion of products. And broken down,AI photo generation or photo editing tools account for the highest proportion, with a total of 13 products., is a product type that developers are relatively fond of and currently seems to be a product type that C-end users are willing to try.
Category 1, AI chat,"Companion attribute" products are more suitable for mobile terminal development
AI assistants and companions are actually chatbots most of the time. There are 9 such products in the Top50.
Head products, such as the top three ChatGPT, character.ai and Google's Bard, only use self-built large models, while several other products are compatible with mainstream language models on the market, such as GPT, Claude, etc., whether they have self-built models or not. Moreover, most products are also compatible with existing Vincentian graph models, evolving into multi-modal AI products. There are also some products such as YOU, Perplexity, and GPTGO.ai that are interested in the search engine business and have received financing. For details, please refer to our previous article "Capital is willing to give money. Now, everyone wants to be a search engine." But overall, everyone is pretty much the same.
Screenshot of the main interface of character.ai and the test results of the author and the AI playing Musk | Image source: character.ai
The only one worth mentioning is character.ai, because it has RPG elements.It can also be regarded as an AI companion, so its DAU on the mobile terminal is the highest among the Top50 products, much higher than ChatGPT.According to Diandian data, character.ai currently has 361W double-ended DAU globally.
Comparison of website and app traffic of various products | Image source: a16z
In fact, in our daily contact with developers, we will find that AI companions are also considered to be one of the most feasible models at present.
Category 2, AI text creation,Japanese users are enthusiastic about using AI to write novels and draw comics
There are only the above four products for AI text creation, which are divided into two categories according to functions and usage scenarios. QuillBot and Smodin are auxiliary writing tools.The usage scenario is more professional and academic, and the main audience is students, teachers, etc.AI-Novel and NovelAI are mainly article generation tools, as can be seen from their names.The usage scenario is the creation of literary works such as novels. While there is some overlap in functionality between the two categories, the overall products are very different.
Overview of QuillBot functions｜Image source: QuillBot
Two AI-assisted writing products have similar functions, here we take the higher-ranked QuillBot as an example. The main functions include grammar checking, paper plagiarism checking, assisted writing, AI automatic summarization, abstract generation, AI translation, etc. Most of them are used in academic scenarios, and are similar to the functions of Grammarly that we are familiar with. Mainly aimed at professional articles such as student papers and corporate reports.
For applications in literary scenes, AI-novel and NovelAI not only have similar names, but also focus on AI automatic writing. Both products come with a text-based drawing function, allowing users to generate illustrations based on the plot at any time, or draw completed stories into comics.
In terms of use, the user gives a start and makes basic settings, and the AI starts to generate text. If the user wants to control the direction of the plot, he or she can write part of the text at certain nodes to guide the AI to continue creating in a certain direction.
Screenshot of AI-Novel website | Image source: AI-Novel
NovelAI style selection function introduction diagram｜Image source: NovelAI
The difference between the two AI novel creation products is that one is a typical two-dimensional style, and the other is biased towards European and American literature. In addition, among the subscription rights of NovelAI, users can choose to let the AI imitate writers of different styles to write, and can also control the plot line and theme of the text through options, while AI-Novel does not have this function. But the risk of imitating different styles of writing is still quite high now. The author of "Game of Thrones" just sued ChatGPT a few days ago.
The results show that AI-Novel has higher visits and a much higher ranking. But in fact, the rankings are very different, one is 16 and the other is 49. In fact, the August visits of the two websites, one is 6.6 million and the other is 4.7 million, the difference is not that big.
And it’s very interesting that 80%+ of AI-Novel’s traffic comes from Japan. NovelAI’s largest traffic source is also Japan. In the field of AI, online articles, and comics, in the mobile Internet era, although Chinese and Korean companies are stronger, in the AI era, Japanese users are very enthusiastic.
Screenshot of Novel interface, traffic statistics｜Image source: NovelAI, SimilarWeb
In addition, online writing is a popular way to go overseas. In addition, the IP chain of the entire content can be used in more and more links, such as online writing, radio dramas, short plays, etc., and overseas manufacturers with massive IP and literary works are also exploring the path of AI-assisted creation.
Category 3. AI audio and video:Editing tools are more popular than generation functions
AI audio and video products, note: red boxes are AI audio products, yellow boxes are AI video products
Among the Top50 products,There are 3 AI audio products in total, Speechify and ElevenLabs are AI reading tools. Users upload text materials and a realistic AI voice will read them. You can choose various styles of voices and also support multiple languages. The two products are relatively similar in terms of functionality, but Speechify does a better job on the mobile app, with a global dual-end average DAU of 2.4W in the past 30 days, while ElevenLabs does not have an official app on the shelves (there is an app with the same name, but it was observed that it is just a shell application that calls the ElevenLabs API).
In terms of effect, we tried Speechify. The speech rendering effect has been greatly improved compared to before, and the English effect will be better.
Screenshot of VocalRemover interface | Image source: VocalRemover
The scope of use of VocalRemover is smaller, mainly in music. As the app's name suggests, it can separate vocals from music to generate accompaniment. In addition, it also has higher-level functions of separating vocals, drums, and musical instruments. In addition, it also has the functions of a normal music editor, including merging tracks, editing, transposing, etc. It is the only AI music product on the Top50 list.
Summary of AI video tool functions
There are 6 AI video tool videos in total, Overall, video generation and video editing are the two main functions. The main differences between different products are the style and usage scenarios of generated content.
runway’s video editing function｜runway
Kaiber video conversion function｜Image source: Kaiber Gallery
Kaiber and runway are geared toward artistic scenes, and their main recommended function is the AI-generated video function. The overall user interface is not much different, but Kaiber will have more options and can more accurately generate the videos that users need. In terms of painting style, based on the works of other users, Kaiber is more animation style, while runway is more realistic. In addition, runway has some video editing functions such as background cutout and image quality restoration, while Kaiber puts more emphasis on the function of converting realistic painting style into animation.
In addition, runway has also launched RunwayML on the mobile terminal, which can use video generation functions similar to those on the web page, but cannot use video editing functions. The global average DAU for both terminals in the past 30 days has been 3.5W.
The remaining four products all have relatively clear business scenarios., except VEED.io, all have both video generation and video editing functions, but the main functions and usage scenarios of each product are different.
Among the four products, two are mainly promoting the video generation function, namely Fliki and D-ID.
Virtual speaker generated by D-ID｜Image source: D-ID
The main function of D-ID is to generate AI digital people. Users only need to upload a photo to generate an AI digital person that can give a speech. Ordinary users can only generate speech videos of the digital person, but enterprise users can call D-ID's API to realize the live broadcast of the digital person. AI digital humans have become relatively mature in China. Many companies such as Xiaobing Technology and Meitu have launched corresponding products, which have also been used in fields such as live streaming and delivery. However, at present, they are mostly used as substitutes for real people.
Fliki’s video generation function (translated) | Image source: Fliki
Fliki, on the other hand, seems to be a more familiar text-based AI video generation tool. Users can not only enter text, but also convert social media content or PPT into videos. It mainly serves enterprises and creators and is used in commercial scenarios such as social media marketing, corporate introductions, product demonstration videos, and training videos.
VEED.io video editing interface｜Image source: VEED.io
The two products KAPWING and VEED.io are actually more like traditional video editors., AI only participates in some functions such as intelligent editing, translation, subtitles, script generation, and background deduction. Only KAPWING has a video generation function, and its scope of use is mainly to generate summary short videos. VEED.io has no AI video generation function and is a purer editing tool.
In terms of traffic, VEED.io’s traffic peaked at 15 million visits in August, while the other three products had only 1/3 or less of VEED.io’s visits. It can be seen that although products like runway are "famous",Commercial AI video direction, auxiliary editing and other more down-to-earth functions have higher user demand.
Category 4. AI Pictures: Pictures and AIStill the most"chemical reaction",Chinese developers win the “ticket”
Image-oriented AI products. Note: The red box is a picture generation product, the yellow box is an AI image processing product, and the blue box is a product with both functions. Note: This part will introduce runway’s AI picture-related functions, so it is also marked in the picture.
There are 14 products specifically for pictures in the Top50, accounting for the highest proportion among all product categories. One can imagine the fierce competition in the field of AI pictures, but at the same time, the fact that so many products have entered the list also shows that the demand does exist.
When we observed AI image products, we divided them into image generation and image processing categories. Among all products,There are 9 products with only image generation function, 4 products with both functions, and only 1 product with photo editing function but no photo generation function, PhotoRoom (but strictly speaking, PhotoRoom also has its own AIGC function, but it is more for commercial purposes, such as generating backgrounds based on text descriptions).
PhotoRoom function list and background transformation function demonstration｜Image source: PhotoRoom
And enter the Top 50 AI photo editing websites,An obvious difference from the mobile terminal is that although both are open to the public, they are more commercial-oriented.Taking the top-ranked PhotoRoom as an example, it basically targets small B businesses. When the author uploads a random picture and PhotoRoom removes the background, what appears is a full screen of template images, suitable for different categories (such as shoes or clothes, electronic products or daily necessities), different social media platforms, different occasions, etc. The functions of the web page should be more complete.
In addition, a very interesting point is that although PhotoRoom ranks 6th on the list, which is already very high, according to a16z data, its main traffic entrance is the mobile terminal, accounting for more than 80%.This shows that although it is a business scenario and everyone thinks that this kind of work will be done on the web, the data shows that the proportion of small businessmen or creators who use mobile phones to take pictures and then directly use their mobile phones to edit the pictures is very high.If readers want to learn more about PhotoRoom, please refer to Beluga Chuhai's article "Where is the way out for photo editing apps amid the AIGC trend?" PhotoRoom demonstrates how monthly revenue can increase to over 10 million."
ZMO.AI function list｜Image source: ZMO.AI
4 Apps with both AI photo retouching and photo generation functions, in terms of targets and users, it also covers both B-side and C-side. Two of these four apps come from China. The 17th-ranked cutout.pro is headquartered in Hong Kong, China. According to information on Linkedin, the developer is named LibAI. They have also developed another AI picture product, promeAI. ZMO.AI, ranked 40th, is a company in Hangzhou. According to current information, only these two products among the Top 50 are from Chinese developers.
Products with both AI photo retouching and photo generation functions: PIXLR, cutout.pro, runway function summary
Image generation product traffic summary in August｜Data source: SimilarWeb
For the image generation app with the largest number of products, the difference in functions is small, and the difference lies in the specialization and style of various drawing models. At present, Midjourney is still the product with the highest traffic, but comparing the a16z list and SimilarWeb data, the traffic gap between various products is decreasing.
Midjourney traffic trend｜Image source: SimilarWeb
Looking at the trend, Midjourney’s traffic data has dropped significantly.Within half a year, the number of visits to the website dropped by half, and this traffic seemed to have flowed to some more niche photo-generating products.For example, Leonardo, which focuses on generating game illustrations, currently accounts for 75% of Midjourney’s traffic.
Comic style sample picture (left) and fantasy style sample picture (right) | Image source: Playground
Judging from several mid-waist products, the styles of pictures generated by these products are mainly fantasy and comic styles. Since the largest user groups of these products are from the United States, this should be the style of painting that American users prefer. However, there are also some products that enter the market through special painting styles or functions.
PIXAI interface (top), traffic distribution data from June to August 2023 (bottom) | Image source: SimilarWeb
For example, PIXAI, ranked 47th on the list, is a platform mainly based on the two-dimensional style. It had 8.47 million visits in August, firmly ranking in the second echelon, and the number of visits in the Japanese market has increased significantly.
DeepSwap traffic trend｜Image source: SimilarWeb
In addition, DeepSwap, ranked 50th on the list, is the only AI face-swapping product on the list. It is a relatively special product type among the mainstream models of text-based image generation. And the recent traffic growth has been very obvious. Although AI face-changing has certain controversies at the legal and privacy levels, demand has always existed among users.
DeepSwap interface｜Image source: DeepSwap
In addition to the AI face-changing function, this product also has simple picture generation (sexy female pictures) and picture editing functions. Except for picture generation, which is free, other functions require payment. The price is $19.99 per month or $99.99 per year (half price for the first payment). The overall price is relatively high.
In fact, after taking stock, we can find that the entire generative AI track is still in a very preliminary stage. Except for ChatGPT, which is difficult to shake, the gap between other products is not very big. For non-headline products, the more specialized you are in a field, the easier it is to enter the market and gain a firm foothold. From a user's perspective, products with practical functions, such as improving work efficiency and enhancing the attractiveness of "works", account for a high proportion, while products with "fantasy" functions but fewer practical scenarios have a downward trend in traffic. The AI picture track seems to be the only track where Chinese developers have mastered the “admission ticket”. Will any new players enter the game in the future, and how will they enter the market in what areas? We will continue to pay attention.