Hypit turns reference clips into code instead of finished video

Hypit converts reference videos into editable video projects by pairing with AI agents such as Codex and Claude Code. As woshipm details, the open-source project operates through a five-step workflow. It begins with inputting assets and analyzing the reference video. Next, it drafts an editable video project and runs Hypit Runtime before saving the project for reuse. Other tools pursue that same structural control. AICRON links 150 AI models into a node-based canvas workflow, including Seedance and Kling alongside GPT Image and Veo.

Video production is moving past one-shot text prompts. Real workflows require an editable foundation where decisions survive revisions.

According to geekpark, Jianying Professional Edition introduced Jianying Hub (剪映 Hub), combining an infinite canvas with a multi-track editor. Creators draft storyboards on the canvas and call platforms like Jimeng or Xiaoyunque for visual and audio assets. They then drop rough cuts into a timeline. AICRON takes a parallel path, and the platform provides over 260 AI functions. These tools handle image and video manipulation alongside audio processing. They also support text generation and 3D visual capabilities. These project layers preserve structure so creators do not have to rebuild clips from scratch.

Inside the project layer that outlives the generation model

Hypit approaches video creation as code rather than an irreversible render. According to woshipm, the open-source project connects with AI agents such as Codex and Claude Code to turn reference videos into editable projects. It builds structure first. Before generating any media, the system constructs a three-part draft containing a Brief and a Treatment shooting outline alongside a main.svml file. Hypit uses this SVML description language to store dialogue and visual assets in a single source file. Subtitles and effects are preserved in the same structure.

Structure outlives the render. Creators can alter underlying instructions without regenerating an entire sequence from scratch.

As woshipm explains, this separation anchors Hypit's five-step workflow: inputting assets, analyzing the reference video, drafting an editable video project, executing via Hypit Runtime, and saving the project for reuse. Other environments organize this layer spatially. According to geekpark, Jianying Professional Edition introduced Jianying Hub (剪映 Hub) as a workspace uniting an infinite canvas and a multi-track editor. Creators draft scripts and storyboards on the canvas before calling generation platforms such as Jimeng and Xiaoyunque to produce visual and audio assets. They then import rough cuts into the timeline.

Spatial systems also coordinate external models. AICRON operates a node-based canvas workflow incorporating 150 AI models, including Seedance, Kling, GPT Image, and Veo. The platform provides over 260 AI functions covering image and video workflows. Its capabilities extend to audio and text, with separate tools dedicated to 3D visuals. Every choice remains modular. Because each tool occupies an independent node, the project remains an open schematic rather than an unalterable export.

Binding assets to words to survive script revisions

Hypit operates by analyzing a reference video to extract sentence-by-sentence timelines and camera cut points before generating a new video, according to woshipm. Standard editing software pins clips to static timeline timestamps. Hypit breaks that constraint by attaching voice tracks and subtitles directly to spoken dialogue lines, and images and animations also bind directly to specific words in the dialogue. Timing adjusts automatically whenever the script is modified, preserving the pacing of the source material.

Editing dialogue in Hypit's SVML file automatically realigns connected audio. It also automatically realigns connected visual elements. This layout allows local re-rendering without regenerating media assets.

Production efficiency improves when rough cuts lock to speech transcripts rather than timeline markers. In a demonstration at a launch event reported by geekpark, Jianying Assistant condensed a 6-minute talking-head raw clip into 3 minutes in 10 minutes. That automated task required an AI point cost of approximately 2 yuan, compared to 32 minutes for manual editing. geekpark also cited Jianying's claim that creating a commentary video for a two-hour film can shrink from approximately two and a half hours to 40 minutes using its tools.

Even with synchronized project layers, structural automation hits operational boundaries. woshipm contributor Mo Yu Xiao Li argues that Hypit cannot achieve fully automated one-click video production because human intervention remains unavoidable. Editors must step in to correct repetitive host gestures and troubleshoot generation timeouts. They also have to polish uneven AI-written dialogue quality before publishing.

Model neutrality against the closed timeline

On September 20, Jianying held the AI New Creation Conference to update its Professional Edition and mobile application, alongside revisions to its creator ecosystem. As geekpark reported, the suite anchors users through an AI creation agent called Jianying Assistant (剪映助手), branded as Xiaoying (小映) on mobile devices. Jianying tied that agent directly to a closed economy. It launched an AI Ultra membership priced at 199 yuan per month for continuous auto-renewal, marked down to an initial launch price of 99 yuan. That subscription supplies 22,100 AI points per month alongside Jianying SVIP benefits.

Direct timeline control keeps creators locked within one proprietary editor. Studio workflows often refuse that boundary.

Founder Lee Soo Young brought more than 20 years of experience in the traditional film and television industry to a model-neutral workspace. Rather than restricting artists to internal generation tools, AICRON has integrated 150 AI models, including Seedance, Kling, GPT Image, and Veo. The software acts as an orchestration surface instead of an isolated timeline.

Production economics reflect that divide. AICRON derives 50% of its revenue from AI film content production and 50% from tool subscriptions, having produced the sequence in Agent Kim: True Colors Return. To coordinate pipelines, it launched AICRON Teams, a collaboration tool featuring a team dashboard that tracks token and credit consumption. While Jianying uses points to protect its native timeline, AICRON treats outside engines as interchangeable utilities managed from a single desk.

Selecting AI Video Production Frameworks and Tooling

• You need script-driven, programmatic video editing where modifying dialogue automatically adjusts subtitle and visual timing without re-rendering every asset from scratch. Use Hypit. Its SVML format binds subtitles, audio, and visual elements directly to spoken dialogue lines rather than fixed timeline timestamps, enabling local re-rendering when script copy changes. Generating test projects has cost between $0.17 and less than $2 in external model calls.
• You require broadcast-grade narrative sequences combining diverse AI modalities in a visual node-based workflow for film and television pipelines. Use AICRON. Created by production company MORPHEUS and proven in broadcast sequences for the Korean drama Agent Kim: True Colors Return, it connects 150 AI models (including Kling, Veo, Seedance, and GPT Image) across more than 260 AI functions on a node canvas.
• You are creating long-form commentary, talking-head cuts, or multi-branch interactive video narratives within an integrated suite. Use Jianying Professional Edition with Jianying Hub and ICG Studio. The workspace connects an infinite storyboard canvas directly to multi-track editing, cutting tasks like a 6-minute raw clip down to 3 minutes in 10 minutes for approximately 2 yuan in AI points, while ICG Studio facilitates branching interactive story creation.
• You manage a creative team and need centralized monitoring of token and credit expenditure across multi-model asset generation. Use AICRON Teams. It provides a collaborative team dashboard specifically built to track credit and token consumption across members, operating on a credit-based tier structure.

Accounting for review time and rerun fees

Hypit itemizes expenses differently. Before paid generation begins, the software displays required external services alongside their projected costs. Producing three official English sample videos cost approximately $1.10 each. A Shiba Inu test video cost less than $2, pairing transcription priced at a few cents with free text-to-speech. Earlier asset passes cost even less. Rendering the storyboard stills and character look test for a Zootopia-themed project required $0.17.

Granular billing matters because single frames require isolated intervention. When generation fails, regenerating entire sequences wastes money. Inspection preserves the budget.

Hypit links inspection to the timeline: selecting a frame highlights its specific line of code, and updating that text refreshes the visual preview. That precision addresses persistent pipeline failures. Contributor Mo Yu Xiao Li writes for woshipm that true one-click production remains unworkable. Human operators must fix generation timeouts and repetitive host gestures. They also have to repair AI-written dialogue.

Trade-offs also extend to infrastructure. A woshipm author notes that free local generation tools produce worse visual results than paid hosted services like Seedance. Configuring that stack has built a secondary market on Xianyu. Merchants sell Hypit project materials for 95 yuan and offer remote installation for 9.9 yuan. Others market proxy deployment or debugging services for 30 yuan and 100 yuan.

Comparison of AI Video Project Layers: Hypit, Jianying, and AICRON

Look closely at what remains editable after the first render. In Hypit, an SVML source file stores dialogue alongside visual assets, while subtitles and effects are saved in the same file. Editing a spoken line automatically realigns connected elements. Editors can re-render scenes locally without regenerating media assets because animations attach to spoken dialogue rather than static timestamps.

The production costs make rapid testing feasible. Hypit's three official English sample videos cost approximately $1.10 each to produce, while storyboard frames for a Zootopia test cost $0.17.

Even so, these files produce rough cuts rather than finished work. As woshipm contributor Mo Yu Xiao Li argues, human intervention remains necessary to fix repetitive host gestures and resolve generation timeouts. AI-written dialogue still requires manual cleanup. Treat the code as a baseline structure, but keep editorial judgment at the center.

For readers outside China

• Availability: Hypit is open-source and globally accessible via GitHub (https://github.com/hypit-ai/hypit) and npm installation via command 'npx skills add hypit-ai/hypit -g' for agents such as Claude Code, Codex, and Cursor's Agent. Third-party installation and materials are also traded in China on Xianyu. Jianying Professional Edition and Jianying Hub were launched on September 20 for the Chinese market, closely integrated with ByteDance domestic ecosystems (Jimeng, Xiaoyunque, Fanqie Novel, Douyin); overseas availability for Jianying Hub and ICG Studio is not disclosed in sources. AICRON is operated by South Korea's MORPHEUS and NextPlay Entertainment.Ltd; international access policies outside Korea are not disclosed in sources.
• Pricing: Hypit itself is open-source, but requires external generation services: character look tests and storyboard stills cost $0.17, a Shiba Inu-themed test video cost less than $2, and official English test videos cost approximately $1.10 each. Unofficial setup and materials on Xianyu cost between 9.9 yuan and 100 yuan. Jianying offers an AI Ultra membership priced at 199 yuan per month for continuous auto-renewal (launch price 99 yuan), providing 22,100 AI points per month. AICRON's mid-tier subscription costs $59 per month for 4,500 credits, which can produce approximately 30 minutes of video.
• Closest Western equivalents: Descript; ComfyUI; Runway; CapCut
• Data residency: Specific server locations, data residency commitments, and international regulatory compliance frameworks for Jianying and AICRON are not disclosed in sources. Hypit executes project rendering locally via Hypit Runtime, but it routes requests to third-party hosted APIs for speech-to-text, dubbing, image generation, and video generation, subject to the data policies of those independent model providers.

Sources

• geekpark 剪映，杀入 AI 互动影游 https://geekpark.net/news/370883
• 36kr AICRON参与《金特务》制作，韩国主流剧集首次引入AI生成内容｜早期项目 https://36kr.com/p/3966062388780290
• woshipm 13K Star开源项目，一键复刻千万网红爆款视频 https://woshipm.com/ai/6468934.html
• woshipm 开源AI视频复刻神器Hypit爆火：闲鱼已有人靠它月入过万？ https://woshipm.com/ai/6469243.html