Video editing has become cheaper, while the capital is flowing into computing power.
How much of the work in a single video can be handed over to AI?
Recently, an editor ran a test, letting Claude Opus 5.5 independently produce a short video.
The AI first called Fish Audio (an AI voice synthesis tool) to generate voiceover in his voice, then used HeyGen (a digital human video generation tool) to drive the digital avatar to speak. The truly complex part was the subsequent editing.
Before getting started, the AI analyzed two reference videos frame by frame, and summed up three rules: every sentence has a corresponding visual scene, every movement is aligned with the beat, and the entire video runs through a single narrative thread. Later, it wrote these rules into codes, automatically completed dynamic effects, configured sound effects, and encapsulated the whole set of methods into a reusable Skill.
Next time, creators only need to provide new voiceover, and the AI can produce videos according to this set of methods. Even if it is replaced with real-life shooting footage, the AI can automatically select materials, extract characters, and place text and dynamic effects in corresponding positions.
According to this editor, no key frame in the whole process was produced manually.
In the past, editors and post-production special effects personnel needed to spend a lot of time learning complex software operations to turn ideas into pictures. Now, AI can not only replace humans to complete operations, but also start to learn the creative methods of editors.
An editor engaged in AI editing told *Dingjiao One* that they once recruited a group of human editors as reference to compare the editing effects of humans and AI. The original goal was to make AI as close to human level as possible, but it turned out that “in some cases, AI edits more coolly than humans”. In her workflow, it is commonplace to use AI to complete tasks with professional software such as Photoshop and Premiere.
Technological changes are also reshaping the competitive landscape of this industry.
In September 2026, Jianying launched the AI creation workstation Hub, intelligent editing assistant and AI Ultra membership, extending its business from editing to conception, generation and delivery. Adobe is also promoting the agent capabilities of professional software such as Premiere. Companies including Descript and Runway cut in from their respective strong links, and extend to a more complete video production process.
In the past, editing software competed for functions and operation experience. Today, models, tools and content platforms are competing for the production process of the same video. The money originally paid to editors and software companies begins to flow to models, computing power and automated services.
01. Editors’ working hours have been converted into AI credits
Editors’ time is often consumed in a string of tri
Take a two-minute product promotion video as an example. The editor downloads the materials, imports them into Premiere, watches them one by one, and separates the voiceover, product close-up and environmental shots; then cuts out misstatements, pauses and repeated expressions, arranges the shots according to the script, and matches corresponding auxiliary pictures for each sentence.
Rough cut is just the beginning.
In order to highlight the selling points, the editor also needs to open Photoshop to cut out the product from the background; switch to After Effects to make title animations; then go back to Premiere to add subtitles, background music and transitions, adjust key frames frame by frame to make the pictures align with the beats. Exporting does not necessarily mean the end. If the client thinks the opening is not attractive and the product appears too late, the editor has to reopen the project, modify the shots, subtitles and dynamic effects, and export again. A seemingly simple adjustment may involve several software and multiple tracks.
In the past few decades, digital creation tools have been evolving, but the division of labor between humans and software has not changed fundamentally. Software is responsible for providing tools, and creators are responsible for performing operations. Whether it is mechanical material sorting or creative picture design, people need to convert ideas into a series of specific operations.
AI is changing this division of labor.
In an AI workflow with capabilities of material identification, model calling and software operation, creators can directly put forward requirements: for example, compress the six-minute voiceover to three minutes, delete repeated expressions, match pictures for each selling point, and add subtitles and dynamic effects according to the specified style. The AI transcribes the voiceover into text with time codes, filters materials and arranges shots, and calls other tools to complete matting, dubbing and special effect production when necessary. After the first version is generated, creators put forward modification opinions through natural language.
Tasks that previously needed to be operated item by item have begun to become work that can be directly delivered to software.
In January this year, Justine Moore, partner of Silicon Valley venture capital firm a16z, proposed that video agents may change video production just like Cursor changed programming. She believes that 2025 is the year of AI video, and 2026 will be the year when people hand over editing work to agents, “AI editing agents will dramatically improve the quality of all the videos we see and speed up their creation rate in the coming months and years”.
There is already data to support this change.
In the demonstration at the Jianying press conference in September this year, the 6-minute original voiceover material was edited into a 3-minute finished film, manual rough cutting took 32 minutes, while Jianying Assistant only took 10 minutes, with a credit cost of about 2 yuan.
According to the customer case announced by AI editing tool Spingle, American documentary production company FlowVideo used to spend four hours looking for auxiliary shots corresponding to the interview content from a large number of materials, and then spent six hours completing the rough cut. After using Spingle, the two tasks were shortened to half an hour and three hours respectively. The total production working hours of the whole project were reduced from 20 hours to 13.5 hours, down by about one third.
For institutions that need to produce videos in batches, efficiency improvement can be converted into greater production capacity, and even reduce part of labor expenses.
According to the customer case data released by OpusClip, American marketing agency Tribe Consulting used to pay about 500 US dollars a week to hire editors to produce 3-5 short videos. After introducing OpusClip, its weekly video output increased to 10-20. Another marketing agency, Full Send Marketing, said that after using AI to complete basic editing, subtitle adding and auxiliary shot matching, the company saved about 1500 US dollars per month.
The above data comes from the demonstrations or customer cases disclosed by tool manufacturers, which does not mean that all video projects can achieve the same range of cost reduction. But in some tasks with high standardization, it has been a trend that AI begins to undertake the work that previously needed to be completed by humans.
For software companies, this means a new charging opportunity.
In the past, video production companies bought software and then hired editors to complete operations. Software is usually charged according to the authorization period or user seats. No matter an editor completes 10 or 100 videos a month, the income of software companies will not increase synchronously accordingly.
Now, when software can directly perform tasks for users, it has the opportunity to charge according to the amount of work completed. Part of the work that originally needed to be completed by adding more staff or investing more working hours can be solved by purchasing AI services.
In September this year, Jianying launched the AI Ultra membership, with a monthly continuous subscription of 199 yuan, including 22100 AI credits per month, and packaged the original editing membership rights and interests.
Descript adopts a similar mixed billing method. Its Creator package is 35 US dollars per month for monthly subscription, including 30 hours of media processing quota and 800 AI credits per month. What users pay for is not only the right to use the software, but also a certain quota of AI processing and generation capabilities.
For enterprise-oriented video processing services, the billing method is more refined.
The short drama highlight intelligent editing service of Volcano Engine charges according to the duration of input materials, with 2.5 yuan per minute for AI analysis of the original film, and video synthesis, media asset storage and distribution are charged separately according to re
From charging by software authorization or subscription cycle to adding billing by material duration, generation times and AI credits, a new price system is emerging in video production.
02. Four groups of players are competing for the same production line
AI editing has brought new revenue opportunities, and also attracted companies that were originally in different industrial links.
At present, there are mainly four groups of players.
The first group are platform ecosystem companies such as Jianying and Kuaiying.
Jianying relies on Douyin, CapCut relies on TikTok, and Kuaiying is backed by Kuaishou. In the past, they lowered the creation threshold to allow more users to produce and publish videos, bringing continuous content supply to the platform. Later, the editing tool itself has also developed into a business.
According to the estimate of mobile application data agency Appfigures, in August 2025, CapCut’s monthly net income on Apple and Google app stores worldwide reached 93.5 million US dollars. This figure does not include the PC-side and web-side businesses.
Now, Jianying hopes to extend its business scope from editing to the entire video production process.
The newly launched Jianying Hub integrates script planning, AI material generation and multi-track editing into the same workstation; Jianying Assistant enters specific projects to undertake work such as voiceover deletion, subtitle error correction and video packaging. In the past, creators might first use large models to write scripts, then call video models to generate materials, and finally return to Jianying to complete post-production. Now, Jianying hopes to keep these links in its own product system as much as possible.
Compared with independent software companies, platform-based players have another advantage: they can connect video production with content distribution and advertising delivery. For example, the Chuangzao platform under Ocean Engine has integrated AI scripts, digital human finished videos and intelligent editing into the advertising creation process. Merchants can use product information and existing materials to produce videos of different versions in batches for subsequent delivery tests.
The second group are professional software companies such as Adobe and Blackmagic.
Film and television production involves material management, multi-person collaboration, special effects, color grading and version delivery. The project file of a film needs to be circulated repeatedly among different teams. The long-term accumulated advantages of professional software are not only functions, but also a production system that can stably complete complex projects.
Therefore, these companies choose to let AI enter existing software, rather than abandoning the traditional timeline and project files.
In June this year, Adobe announced that it would extend Creative Agent to products such as Premiere. Users can sort materials, identify interview content and generate rough cut sequences through natural language, and the results completed by AI still remain in the editable project, for creators to continue modifying.
Blackmagic opens up to external agents. DaVinci Resolve 21.1 released in September added integration capabilities with AI assistants such as Claude, Claude Code and Codex, allowing external agents to analyze projects, sort materials and render in batches.
The two companies have different routes, but similar goals: even if AI undertakes more and more specific operations, professional production still cannot do without the underlying capabilities they provide. What they sell is still dominated by software subscriptions, authorizations, professional versions and related products and services, and the primary task of AI is to improve product value and retain customers.
The third group are AI-native companies such as Descript, OpusClip and Sparki.
Different from traditional software, this type of company hopes that users do not need to learn complex operation methods, and only need to tell AI what they want.
Descript allows users to edit videos just like modifying documents; OpusClip focuses on automatically cutting long videos such as podcasts and interviews into short clips; Sparki allows users to upload materials and describe their goals, and the system organizes shots, subtitles and music.
These companies do not need to take into account the huge historical functions of traditional software, can redesign products around specific tasks, and make money through subscriptions, usage quotas and pay-as-you-go billing.
But they face a problem: how high a premium can independent AI editing capabilities form?
On the one hand, Jianying and Adobe are integrating similar functions into their existing products; on the other hand, general agents have begun to have the capabilities of understanding materials, calling software and performing editing tasks. For users who have already purchased editing software or general large model membership, there is no reason to purchase an additional AI editing product unless the advantages in quality, efficiency or cost are sufficiently obvious.
The fourth group are upstream video model companies such as Kuaishou Keling and Runway, which are entering post-editing from material generation.
In the past, video generation models were mainly responsible for material production, and editing software was responsible for combining materials into complete works. Now, the boundary between the two begins to blur.
Runway launched Aleph 2.0 and Edit Studio in May 2026, allowing users to modify specific objects, backgrounds and picture styles in existing videos while retaining other content as much as possible. Video model products such as Keling are also expanding their capabilities of shot control, video modification and local editing.
When models begin to have editing capabilities, they will no longer only provide materials to editing software, but have the opportunity to directly undertake part of post-production tasks, extending the charging scope from video generation to video modification and production.
The four groups of players are extending their boundaries from different directions.
Jianying integrates scripts and material generation forward, Adobe lets agents enter professional projects, AI-native companies try to directly deliver editing results, and video model companies such as Runway extend model capabilities to post-modification.
What they are competing for is no longer just a certain editing function, but the control right of the video production process. Who can master the user entry, who can control the key production links, and who can retain profits after paying the cost of models and computing power?
03. Editing is getting cheaper, and it is getting harder to make money
Software and AI companies are talking about efficiency, but for creators, AI editing may not be a cost-effective deal.
A video self-media creator told *Dingjiao One* that she has been used to paying for AI: using Kimi to assist in writing scripts, and then using AI video generation aggregation tools to make materials, spending hundreds of yuan on AI tools every month. But she has no plan to subscribe to AI editing services for the time being.
In her opinion, AI has greater value in creative conception and material generation links, which is worth paying for; while post-production work such as arranging shots, adding subtitles and special effects can be completed by herself, there is no need to add an extra expense.
Another AIGC creator once tried to let ChatGPT operate professional software such as Photoshop, but finally gave up, the reason is “too expensive”.
The more troublesome thing is that even if users are willing to pay, independent editing tools may not be able to maintain the existing price for a long time.
In October this year, Gen Furukawa, a creator of marketing company SuperMarketers, shared an attempt. He originally paid about 25 US dollars a month to subscribe to Descript for editing interview and voiceover videos. Later, he found that Claude Code can call tools such as Whisper and FFmpeg to complete work such as deleting misstatements, compressing pauses, and adding subtitles.
He encapsulated this process into an open-ls next time. According to his estimate, this can replace most of his work on Descript, saving about 3
