Take Away:

1. The key to Talkie's success lies in the use of multimodal technology to create an immersive AI virtual social experience, and seize the market demand centered on the domestic girl game user group.

2. Talkie's commercial revenue is not the North Star indicator, but the acquisition of user volume and data is.

3. Products that blindly imitate Talkie are likely to die quickly, and they need to have unique technical strength and user insights. While Talkie does have some content limitations, TipsyChat is a new trending app that could potentially surpass Talkie.

4. This wave of technological changes in AI social networking has just begun, and there is still a lot of room for imagination in models, interactions, and business models.

5. As a UGC community in the AI era, there is no network effect.

Product framework and analysis

Let's first analyze the Talkie App, and then give a framework for everyone to understand. Talkie App:

Product analysis. What is scene construction? For example, the Daling Empress went on a secret visit. I can construct an imaginary scene: the person who read the article but did not forward it was caught, and the Empress took the place of the person and began to interrogate: "You, the one who read the article, why didn't you share it?" This is very similar to playing a word game, Mud.

At the product analysis level, we will do a specific breakdown from several angles to see why Talkie's product design is one step ahead.

Character creation. Let's first look at the steps required for character creation:

1. Image creation: Based on the uploaded pictures and description words, a total of 18 anime pictures can be generated. This is a card making that young people like very much, and you can make cards continuously.

2. Character creation: Here you can generate automatically or manually set each field one by one. There are two interesting points:

   • First, you can set the timbre with a high degree of freedom, choose any 5 timbres, set the proportion of each timbre, and then mix them into your own custom timbre.

   • Second, the dialogue style can specify multiple rounds of dialogue, or you can call it directly from the dialogue template. It is equivalent to few-shot, allowing AI to learn from the dialogue style.

3. Confirm the release: You can return to modify here, or set whether the intelligent body is public or hidden. After confirmation, it can be released.

Compared to Character.ai, the degree of customization in the setting of characters (that is, intelligent bodies) is much higher. At the level of character creation, we can think of a total of three layers:

• These constitute most of the elements of communication between characters and people. The image is greatly enhanced by the Xingnian system (we will talk about it later), and therefore Talkie has made a lot of enhancements in this area.

• Personality is the core part of text communication. Talkie has also customized it in multiple dimensions to help the personality be closer to the user's imagination.

• Timbre is also very interesting. This is Talkie's unique design. In the chat interface, the words replied by the character can be clicked on the play button to download the corresponding voice separately and play it, which will also greatly enhance the user experience.

The above three points, at the setting level, together constitute Talkie's stronger sense of immersion beyond its competitors.The official description of Star Thoughts in the tweet on October 4 is as follows:

Star Thoughts are "memory photos" of the intelligent entity and you. The front of it is a beautiful image of the intelligent entity, and the back is the storyline between you and the intelligent entity. Each Star Thought is unique and records your unique story.

It's just that the current version of the product design is not set according to this logic. In fact, Star Thoughts do not have much information that blends with the dialogue system, or it is too random to be noticed. It is also uneditable and automatically generated by the system.

The Star Thoughts system is a unique system created by Talkie. Based on this, the virtual currency in the entire Talkie product can exist. We will spend more time and ink to describe this product design:

The unified entrance is in the personal center, where you can see my Star Thoughts and the Star Thoughts that have not yet been opened.

The so-called Star Thoughts, a blogger of Facebook explained it very simply: Star Thoughts are equivalent to looking at AI photos.

In addition to appreciating Star Thoughts by yourself, you can also download and set them as chat backgrounds, and you can also publish them. Publishing means setting a price to list them for sale in the trading market. After publishing, they will appear in the window of character information.

There is an interesting design in the trading system. The author (that is, the mother called by users) can get a 2% share. This author is the creator of the character, that is, if you create a popular character, you can make money lying down in the future! To a certain extent, it is equivalent to Talkie encouraging users to create attractive characters, because good characters can attract more users to experience them. In order to better encourage, in addition to daily operational activities, a creator certification channel has been added.

There are several sources of Star Thoughts:

1. The first is that in the main thread, after a certain number of rounds of chats between users and characters, a Star Thought will be actively pushed.

2. Users can open Star Thoughts for free once a day, and then they need to spend virtual currency: diamonds to complete the opening. The upper right picture also adds promotional strategies and an entrance to guide the purchase of Star Month Cards. Opening means drawing one card from six randomly generated cards.

There is also the Xingnian Wishing Pool, the entrance is below the character chat interface, and the wishing pool is especially like the card drawing required in many games, which can be drawn individually or ten times in a row. After drawing a card, a card is randomly generated, which can be regenerated 3 times.

Dialogue system

OK, after talking about the above two important elements, let's take a look at the dialogue system that most users can feel. For the dialogue system, Talkie also has many product designs, and we will also explain them one by one.

Let me talk about two details of interaction. First, you can feel it:

The first detail: pay attention to the picture on the right above. When you press and hold the chat interface and slide up, the text in the upper half will be automatically hidden to reveal the background picture. There is also a small button in the lower right corner to quickly return to the latest message.

Another detail is the playback of voice. We mentioned earlier in the character creation that the character's timbre can be customized. In the dialogue system, the voice will be played in two places. The first is that there will be a play button on each character's dialog box with dialogue content, and the voice content will be downloaded only after clicking it. It will automatically play after the download is completed. The second is that when you swipe left to see a new character in the discovery tab, the character's welcome message will be automatically played.

In addition, there are some details about the share button and screenshot function, which you can experience by yourself.

The control of the dialogue process includes six mechanisms: re-speaking and backtracking, memory and restart, and evaluation and event book. Let's look at re-speaking first. Re-speaking is to re-select the latest reply to the character, which is very similar to the re-generation button of ChatGPT. Click to enter the re-speaking page: you can choose the other four replies generated by AI, or you can choose to rewrite the current reply and manually configure the character's reply. The upper right picture shows the manually rewritten content.

The logic of backtracking is that the current character has entered a repetitive dead loop. No matter what you say, it is a logic. If you can't play anymore and are not happy, you can go back to a previous node. This operation is to press and hold the designated dialogue of the character you want to backtrack to, and select backtracking in the pop-up floating window. Backtracking is a bit like the logic of forgetting, and continuing chatting at a certain node again.

Memory: set the character's name for the user, and the perception of the user's gender, and you can set the conversation style at any time. After setting my name as "Super Yellow" in memory, the character remembers this.

Evaluation: for the latest character's reply, you can directly score it so that the character can adjust its reply content and gradually approach the user's preferences.

Finally, let's talk about an event book. The official definition of the event book is: after establishing a deep bond with a certain intelligent body, your exclusive event book will be opened. It looks like a summary of some kind of phased content. I guess it is a means to help the character remember when there are too many dialogue turns and the context length of the model is exceeded.

All of the above really teaches AI how to behave. Continuing, there are two logics for replying. One is to click the input box to reply directly, and the other is to click the small light bulb on the right side of the input box. AI will recommend three reply words (called "inspiration").

There are two mysteries in the reply words. One is that you can unlock the fourth AI-recommended reply word after opening the Star and Moon Card. The other is that there is Star Thought Management in the lower right corner. After clicking it, it looks like this: Different Star Thoughts represent different character personalities, and correspondingly different AI recommended reply words will be generated. This setting is quite interesting and also enhances the significance of Star Thoughts.

Finally, there is a core of charging: Star and Moon Card. To put it bluntly, it is the logic of subscribing to members. Many game strategies are introduced, such as daily login to receive virtual currency Star Diamonds (a common means of increasing activity in games), recharge with additional Star Diamonds, 5 Star Thoughts opportunities per month, and an additional inspiration option in the dialogue system mentioned in the previous section.

Here we have to look at the gameplay of the Star Diamond currency system. I have mentioned it before, and I will emphasize it again here for the Star Monthly Card: In addition to the Star Monthly Card, you can also pay for Star Diamonds. Once your dependence on Star Thoughts increases, you will fall into a cycle of constantly recharging to buy Star Diamonds, and Talkie provides a very game-like mechanism such as ten consecutive draws to provide users with a mechanism for large-scale consumption.

We will talk about this in the commercialization module.

OK, in the previous section, Talkie's overall product was sorted out. In this section, we will do some abstraction and try to further understand Talkie:

Multimodal Combination of Product Design to Create Immersion We can see that Talkie, backed by the large model of Anthropic, has introduced three mature modes of text, voice, and image to jointly create the immersion of the product. This sense of immersion is reflected in the dialogue system: by controlling the dialogue options (six mechanisms: retelling, backtracking, memory, restarting, evaluation, and event book), the process can be maximized by the user, supplemented by a rich customized sound enhancement experience, plus a large number of replaceable background images to build a strong sense of immersive dialogue. Additionally, some interactive details ensure the effect of the realization of the meticulous. Here, the most core text generation, because it is placed in the dialogue system, the ability limitations of the large model itself will be better eliminated (that is, hallucination is not a very serious problem for AI dialogue). Users themselves also know that AI is not omnipotent and can accept the jump of dialogue to a certain extent, not to mention that Talkie provides a high-level means of controlling dialogue.

Based on the sense of immersion, a trading system is derived: If we compare c.ai and other weak domestic products, we will find how excellent the product interaction design of Talkie and Talkie is (although the mobile Internet has developed to a mature point, seeing the ugly interface of the competing products, I am more aware that in the AI era, the mature product methodology has not been migrated in time).

Around character creation, Xingnian, the embedding in the dialogue system, and even the rights and interests of the Xingyue card, the AI photos that can be generated and obtained are an important part of creating immersion. Based on this thinking, there is a whole trading system behind Xingnian. Let's simply go through it again:

Based on our limited observation for about a week, it is judged that the main force of Talkie users is:

• The core is domestic otome game players, spreading to the pan-two-dimensional user group.

• The users are relatively young and sensitive to prices. The core users and otome games have a high overlap. This conclusion comes from our communication, interaction, and content observation from official groups, user groups, ins, facebook, and tiktok.

For this type of user, there are several conclusions about their use behavior in Talkie:

• Properly stimulated creative desire to DIY create their own characters. From image, personality, voice, dialogue style, they can shape them one by one according to their own preferences.

• Pursuit of recognition:

  • Emotional comfort given by the character, this is particularly deep: It is so serious that they call the AI characters they molded "babies," call themselves "baby moms/dads," and strongly condemn the abuse of babies they see (killing and abusing AI characters). It can be compared to the anger of dog lovers during the nationwide dog hunting a while ago.

  • Interactive sharing between players: Share the quotes, stories, and star thoughts of the characters they made, hoping to be recognized by others.

  • Powder promotion: Hope to see others recognize or use the characters they made, and buy the star thoughts they made.

  • Pursuit of visual experience: Card drawing is just what they want.

This part is limited by time and experience, and we are still not doing enough. The above are some preliminary conclusions. What I want to emphasize more is that product design needs to respond to user needs and can exceed user needs, but the premise is to understand the user.

So, what is the positioning of Talkie?

We all know that crush on was the white moonlight in the hearts of users before, and it is also a small hit App. Is Talkie a substitute for crush on?

Talkie as a Substitute for Crush on?

There is a post on Facebook that describes the relationship between Crush on and Talkie:

Talkie and Talkie are both positioned as AI social software, not games! Although it looks like a game in form, it is not a game! Therefore, no agent is needed.

In March this year, Xiyu launched a product called Inspo (artificial intelligence assistant with writing, information acquisition, advice, etc.).

In June this year, the overseas Talkie developer also launched a product called Inspo with exactly the same functions.

At the same time, a timeline was issued:

• In March this year, crush on was launched, a product called Inspo (artificial intelligence assistant, capable of writing, obtaining information, providing advice, etc.). In June this year, the developer of overseas Talkie also launched a product called Inspo, which has exactly the same functions.

• In March this year, crush on was removed from the shelves due to malicious reports. Xiyu went overseas and relaunched crush on. During the overseas rebirth, crush on continued to be rotten.

• On June 16, Talkie was officially launched in North America, which is actually an upgraded version of crush on.

• On August 27, Talkie ranked Top 5 in the total list of non-game downloads on Google Play in the United States and ranked sixth in the total list of downloads.

• On August 31, Anthropic obtained domestic large model registration and was allowed to open services to the public.

• On September 9, the domestic version of Talkie-Talkie was launched in major app stores in my country. According to gossip, it can be cross-verified that the above information is likely to be true.

So, is Talkie a substitute for crush on? Most likely not. There are multiple factors involved. Let's highlight a few key ones:

First, let's focus on the launch time of Crushon:

October 30th, 2022—not October 30th, 2023.

Keep this date in mind, and now let's look at the launch time of ChatGPT: November 30th, when ChatGPT-3.5 was quietly released.

In other words, crush on had already launched in China a month before ChatGPT.

The First-Mover Advantage:

This early launch helped crush on gain a large number of users in just a few months, leading to:

• Massive user behavior
• Massive user data

User behavior allowed Anthropic to better understand its users, while the user data enabled Anthropic to fine-tune emotional companionship dialogues. These advantages provided Anthropic with unique benefits in the AI + companionship track.

Crushon's Removal and Talkie's Launch:

Unfortunately, crush on was removed from shelves in March due to a report. However, after the Anthropic large model passed domestic filing review, Talkie was launched on September 9th.

Is Talkie a Substitute for Crushon?

Many claim that Talkie is either a domestic substitute for crush on or an upgraded version. If you ask me, I believe Talkie is a variant of crush on. While it might serve as both a substitute and an upgrade, it's still a variant for several reasons:

Different Time Points:

• Last year, there were no complete competitors in China, but now the market is much more competitive. There are many teams (including some I know) planning to enter the market. It's uncertain whether previous establishments will continue to hold.

Evolving Understanding:

• After experiencing crush on's rapid growth, Anthropic has gained deeper insights into this user group and evolved its product to better cater to users or even aim to surpass the original group.

Changing Product Form:

• Advances in technology, such as SD, sound generation, and a push toward multimodality, have led Talkie to incorporate picture generation and sound customization—features absent in crush on. Notably, the star thought system has also been added. As a result, the user group will evolve further.

Tipsy Chat: A New Trending App That Could Possibly Surpass Talkie?

Talkie is a variant of Crushon, not a simple substitute or upgrade. However, Talkie does have some content limitations. If users are looking for a more immersive and unrestricted AI Chatbot that focuses on storytelling, TipsyChat is a better choice. Anyone who opens TipsyChat will undoubtedly be impressed by its product design, especially the stunning large-screen character displays and highly immersive chat experience, which set it apart from other products in terms of attention to detail. The interaction design is simple and smooth, making it ideal for new users. While advanced models require a subscription, its basic features are attractive enough for most users, making it a highly cost-effective app.

Product form is changing. The maturity of the technology itself, such as SD, sound generation, and the pursuit of multimodality, has led Talkie to add picture generation and sound customization compared to crush on, including the more important star thought system. Correspondingly, the user group will also change further. The above means that Talkie is a variant of Crush on.

So what is the positioning of Talkie? Let me continue my thinking below:

Is Talkie a competitor of otome games? Has Talkie snatched users of otome games? Let's first look at what otome is: The word "otome" comes from Japanese, where "乙" (おと) means "second" or "another", and "女" (め) naturally means "girl". In Japanese, "乙女" is generally used to refer to young women, especially unmarried girls. This word is often used in Japanese culture to describe young women who are in adolescence and have innocence and girlish feelings. Otome games are usually designed to attract young female players, focusing on the exploration of romance and relationships, and providing a virtual space where players can experience idealized love stories. For example, Mystic Messenger is a typical otome game. In this game, the player plays a young woman who inherits a TV production company that is about to go bankrupt. The core of the game is to promote the story through interaction with four main male characters. Each male character has a unique background story and personality traits, and the player's choices and interactions will not only affect the relationship with these characters, but also their careers and the fate of the company.

The charm of otome games lies in that they provide a virtual space where players can experience romantic love stories while participating in the creation and development of the story. Such games usually contain exquisite visual art, attractive character design, and deep storylines, allowing players to immerse themselves in a virtual world full of emotions and adventures. Domestic otome games are beautifully designed and have rich side quests, but we can feel very clearly that playing them is a limited experience. The whole game has a clear main storyline. We just have to break through level by level and accumulate strength. You can't change the main storyline of the game. It is not similar to an open sandbox, which can give users relatively little independent decision-making power.

Let's take the mobile phone function in Mystic Messenger as an example: There are messages, friend circles, and recent calls, but all of these are pre-set and can only be selected from limited options. Even calls can only be listened to, and there are no interactive items. Why don't I play otome games? Because I can't take control of the main decisions in otome games. I like high freedom and don't care about exquisite cards and sounds. And these are exactly what AI chat can bring me, so there will be users who are frank about the differences between otome games and AI chat. The completely free development system brought by Talkie makes mothers crazy. Here you can do whatever you want and interact with the "male god" in your heart. From this perspective, Talkie is, to some extent, a complete competitor of otome games. But there are also some relatively weak points, which we will expand on when we have the chance.

Is Talkie a UGC community in the AI era? From the previous product analysis, it can be seen that Talkie has a certain tendency to support creators. This is reflected in the "creator certification application" and "author sharing". Thinking further along this line, we can find that:

Users can communicate directly with characters, and can also create characters. At the same time, the created characters can be "copied" to other users for use. Users can be "content" producers or "content" consumers (of course, "content" here needs to be quoted, referring to the characters created by users, or called intelligent bodies).

Users here have created content and completed "content" consumption. To some extent, Talkie is an AI content community. From this point of view, it will indeed snatch users from facebook and ins. At present, it cannot be said that it will not in the future. Personally, I think that the further evolved AI content community is likely to subvert the traditional content community. Of course, the traditional content community will also gradually transform.

AI content communities are very different from traditional content communities (tiktok, facebook, etc.): In traditional content communities, creators directly produce the final content, and the platform mainly completes the review, distribution, and operation work. The content is recommended to consumers through AI algorithms. We are all very familiar with this set. AI content communities are different. Creators only set role information within limited parameters. The big model relies on role information to respond to consumer content interactions. Here are some interesting points:

1. The threshold for creators to create is greatly reduced, and they do not need to be responsible for the final result. After specifying the parameter information, the big model completes the subsequent work.

2. The AI distribution here only distributes roles, not specific content. Content needs to be generated by the interaction between consumers and roles. Next, the consumption of content has greatly increased the participation of consumers. Consumers poke the role, and the role spits out some content.

3. A very core point that is not specifically shown in the above figure is: in traditional content communities, consumers pay attention to content first, and the content usually contains IP, so consumers can easily pay attention to IP; in AI content communities, IP is mainly presented on AI roles, and the author is more deeply hidden, so it is difficult for consumers to pay attention to the author.

---

What to read next:

9 Best NSFW AI Chatbot Apps of 2025

Why AI Lovers Are Transforming Modern Emotional Connections

Why Emotional Intelligence in AI Chatbots Is the Next Frontier