The recent news about AI video is very thought-provoking when taken together.

Sora is here, there is no explosion, only complaints and unsatisfactory results. The good news is that Ke Ling and Conch breathed a long sigh of relief.

November data from the AI ​​product list shows that the growth rate of visits to major AI video products has begun to significantly converge or even decline. Conch increased by 39.33% year-on-year, Keling website visits decreased by 29.99% year-on-year, and Pika decreased by 48.19% year-on-year. The growth of each company last month can be described as "amazing". Conch increased by 2772.92% year-on-year in October, and Pika increased by 787.65% year-on-year.

According to the report of "Smart Emergence", Byte has internally upgraded Jimeng's product priority and is trying to use new paths to create "Douyin" in the AI ​​era. At the same time, Byte plans to transfer more resources to multi-modal product forms in the future. For example, large models related to visual generation will be optimized around Jimeng’s needs.

At this stage, the AI ​​video track has exposed several problems: the first-mover advantage is not obvious, the market is still in a very early stage, reshuffles often occur, and OpenAI is proud of its dimensionality reduction attack It failed; the absolute technical barriers and segmentation differences have not yet been stratified, resulting in insufficient user stickiness. The debut and new releases are the "highlight moments". As time goes by, the traffic begins to decline; the target users are not clear enough, and the people who are currently using the tool Only the top 10% of the pyramid.

An industry insider told Photon Planet: “We are still in the model-as-product stage. Like general-purpose AI assistants, the biggest problem with AI video generation tools is that the scenarios are too general, and ordinary C-end users simply don’t have them. Mind, you also need to draw cards to generate effects."

Even Bytedance’s Jimeng and Kuaishou’s Keling are currently positioned as “universal tools” and are not linked to their products. The unsaturated investment of ByteDance and Kuaishou has virtually given start-up companies a time window to develop. Once the "immediate dream priority" increases, new variables may come.

The integration of Jimeng, Jianying and Douyin may become one of the paths to speed up the AI ​​version of Douyin. The huge number of existing users and C-end market are beyond the reach of many start-up companies. Prior to this, the new fire of AI had already been sown in Douyin.

Sora initially attracted attention because of the DiT model technology it pioneered. Up to now, all products delivered, including OpenAI, are complete products. In this update, Sora's underlying model capabilities are not amazing, but it has built a complete AI video production process and opened up greater editing rights to users.

At present, AI video is still in the early stages of exploration. We can still see that Sora-like products have two divisible routes. Take the professional route to create the AI ​​version of "Premiere".Or the AI ​​version of "cutting" that takes the popular route.

According to Photon Planet’s understanding, although most of the companies have a “Super Creator Plan”, which invites professional creators to try out the advanced version functions to demonstrate the upper limit of their model effects, the ones that were finally launched on the market Functions still follow the principles of simplicity and speed. In particular, Jimeng and Keling deliberately lower the threshold of use in their product design, so novice users can immediately start inputting prompts to generate videos of a few seconds.

Based on this, making an AI version of "cutting" may be the direction that most players want to work hard.

The latest data shows that Cutting and CapCut will achieve more than three-digit revenue growth in 2024, and their total revenue is approaching 10 billion yuan. At the same time, the global monthly active users of Jianying and CapCut have exceeded 800 million.

The problem behind the scenes that cannot be ignored is that Byte Cutting has never been a solo effort, but the trump card played by "Douyin/TikTok + Cutting/CapCut". With Ji Meng, the loop is once again closed. Ji Meng generates materials, cuts and edits, and then digests them in the Douyin content pool. This is why ByteDance attaches great importance to it.

General-purpose AI video is at a crossroads, and startups are building advantageous features and looking for opportunities to further differentiate. Some functions may attract specific groups of users. For example, video production can be classified into realistic, fantasy and other themes. In turn, users can also enhance the attributes and labels of the product.

Byte also has the right to choose. Doubao is an intermediate state. Dream may not be the final product form, but it can be used as a tool and entrance to attract users to come in and use it. Repeated AB testing during the process can not only provide data reference for training the underlying model, but also hatch more AI products.

Up , AI video attempts to seek traditional and authoritative recognition. Downward, the conduction speed is much faster than imagined.

In the second half of this year, AI videos spread at a viral rate and took over platforms such as Douyin, Kuaishou, and Xiaohongshu, creating several waves of popularity outside the industry. A "Venom" special effect that has become popular all over the Internet comes from PixVerse; the "Mylene Fight" AI video that calms the audience's emotions outside the variety show has gone viral, and the watermark that has not been erased shows that it comes from a conch; after the re-release of "Harry Potter" , AI "Wizard Cat" and "Rap Kitten" became popular in China from domestic social platforms.

Users readily accepted the novelty and impact brought by AI technology, and were even willing to pay for it. For a time, they supported a group of middlemen who earned information gaps in Xiaohongshu and Xianyu.

The strongest terminator is Douyin. AI video creation tools including PixVerse, Conch, etc. have all made a wedding dress for Douyin. The popular AI special effects and templates that have been proven by the market will be launched in the editing industry soon. It is easier to get started and you can publish them with one click.Sound, users will never trace it back to which company it comes from. For example, the "rap kitten" that became popular abroad became a popular one among netizens and became "God bless all the cat food and freeze-dried food", and the relevant cut-out template once rushed to the hot list.

At present, AI has found an intermediate state in Douyin - AI special effects. Material templates such as "AI pets dance like they ate mushrooms" and "Everything can be done with AI wool rolls" are still very popular. As of December 9, the "Wool Roll for Everything" material template showed that nearly 3 million people had used it.

It can be seen that before there was a deep connection between Jiemeng and Douyin, Jianying had already taken over part of the AI ​​video functions. This leads to the overall perception of AI videos by ordinary users from large platforms such as Douyin and Kuaishou.

The same dilemma also exists in AI applications in the form of small programs. Meituan’s “Miao Brush AI”, step-by-step soul extractor and AI music replacement are acceptable for a while, but it is difficult to follow up. dissemination use. It may be difficult for users to remember under what circumstances they opened mini programs, but Douyin and Jianying provide entrances to help them build a mental understanding of how to use them.

Perhaps aware of their own disadvantages, AI tools such as Conch and PixVerse have started to go overseas. But recently Photon Planet has noticed that Conch is also increasing its investment in C and acquiring customers.

The value of the sentence "AI made Bilibili make money first" is still rising. Following the Conch AI Assistant, Minimax once again started a new round of investment on the Conch Video platform. We found that some of the up owners of the film and television section of Station B have received advertisements from Conch. Unrelevant promotional words such as "super picture understanding ability" and "instruction following ability" appeared in the complaints video, which seemed out of place.

Minimax’s idea is actually feasible. Station B naturally has its own spoof and ghost genes, which is suitable for the current AI video model presentation effect. The earlier batch of "Mai Lin Fight AI Synthesis" and "Mai Lin Fight Spoof" videos were streamed from Station B. Judging from the comment area, the audience did not feel uncomfortable, but felt that the AI ​​format was more expressive to their viewing of variety shows. emotions at the time.

In addition to good soil, the up owners of station B are also potential paying users. Some up owners often lack materials when making video screens. Some people will use hand-drawing, animation, etc. to replace them. AI generation will improve efficiency.

Of course, there may be copyright and regulatory issues involved. Recently, the State Administration of Radio, Film and Television has issued a document ordering the cleanup of short videos of AI "magically modified" film and television dramas and strictly implementing the content review requirements of generative artificial intelligence.

Looking back at the history of Kuaishou’s fortune, from GIFs and short videos have evolved into comprehensive short-term video community platforms, which to a certain extent can give some ideas to current AI video tools.

AIGC waveChao Zhong Wensheng Pictures and Wensheng Videos are the first technologies to reach the usable stage. The initial training of AI videos relied on text-based pictures and pictures-based pictures, which consisted of countless still frames to create a moving frame effect of several seconds. At a later stage, the one-step Wensheng video and Tusheng video gradually emerged, which is generally similar to the development trajectory from GIF to short video.

The next new stage is probably from general to vertical, from tools to platforms and ecosystems.

In fact, some startups have realized the limitations of a single "tool" and have come up with the idea of ​​becoming a creator and content community. However, the effect is not very obvious, and most players still focus on the development of new functions. Cultivating a mature community requires time and the accumulation of user scale. Just like the current situation described before, even if users go beyond Jimeng and use startup products, we can still see a large number of videos with "Conch" and "PixVerse" watermarks flowing to Douyin, Douyin, etc. Kuaishou, Bilibili.

Nonetheless, this does not prevent new players from imagining the future of native AI video tools. After all, even Byte itself has to "restart" to find the AI ​​version of Douyin.

From AI music production tools to AI music applications, Suno provides a case for the integration of tradition and AI. Opening Suno will give you a sense of intimacy. If you describe Suno in one sentence, it is roughly a combination of "NetEase Cloud Music + Douyin". Familiar product types virtually lower the user threshold. Relying on long-term usage habits, they can quickly acquire the two attributes of "music software" and "social platform".

AI is reflected in content supply. All music on Suno is generated by AI. The original community content mounted on the tool is naturally transformed into a music list. The Douyin-style design completely breaks the barrier from tool to popular application. Even ordinary users can create theme music by casually taking and uploading pictures and videos.

Compared to music, AI videos are more viral and shareable. The intermediate state mentioned by Byte may have another interpretation: AI applications at this stage can be pure AI native, AI-based transformation of traditional applications, or "the most familiar stranger".