short-video-factory vs tensor-house: AI Marketing Automation Head-to-Head (2026)

Two open-source projects are generating the most GitHub activity in the AI-for-marketing space right now. short-video-factory — a one-click desktop tool for batch-generating product marketing short videos — has climbed to 3,964 stars and was updated as recently as May 18, 2026. tensor-house, a curated collection of Jupyter notebooks and demo AI/ML applications for enterprise marketing use cases, sits at 1,440 stars with an update on the same day.

They solve fundamentally different problems. short-video-factory is a practitioner tool: point it at a product, get a polished short video. tensor-house is an engineer's workbench: reference implementations of ML models for pricing, demand forecasting, recommendation, and campaign optimization that data teams adapt into production systems.

This comparison cuts through the surface-level difference — "one is a desktop app, one is notebooks" — and looks at where each genuinely wins, where each falls short, and which type of marketing team should reach for which.

At a Glance

What short-video-factory Does

Per its own repository description, short-video-factory delivers "one-click generation of product marketing and general content short videos" with "AI batch automatic clipping" in a "beautiful cross-platform desktop tool." The Chinese-language description (一键生成产品营销与泛内容短视频，AI批量自动剪辑，高颜值跨平台桌面端工具) confirms the same capabilities with emphasis on the cross-platform UI quality.

That means the workflow looks roughly like this:

• Feed in product assets (images, descriptions, or footage)
• The AI pipeline assembles clips, adds transitions, and applies automatic editing decisions
• You export short-form video files suited to platforms like TikTok, Instagram Reels, or YouTube Shorts
• Batch mode lets you produce multiple variants in one run — useful for A/B testing creatives or covering multiple products at once

The "high-颜值 (high aesthetic quality)" language in the description is a meaningful signal: the project prioritizes polished output, not just functional output. For marketers who don't have a video production background, that matters enormously.

Who Benefits from short-video-factory

• E-commerce teams running product catalogs who need per-SKU video at scale
• Social media managers who need a steady stream of short-form content without per-video editing time
• Indie creators and small agencies who can't justify dedicated video production resources
• Growth marketers who want to iterate on creative quickly by generating batches of variants

Limitations to Acknowledge

The repository is written primarily in Chinese, which creates a documentation barrier for non-Chinese-speaking users. The project's GitHub page does not publish structured pricing data, a hosted SaaS tier, or third-party integrations as of the research date. Teams that need video to slot into an existing content management system or DAM will need to build that connection themselves.

What tensor-house Does

tensor-house describes itself as "a collection of reference Jupyter notebooks and demo AI/ML applications for enterprise use cases: marketing, pricing, supply chain, smart manufacturing, and more." For marketing specifically, that means implementations of:

• Customer segmentation and lifetime value models
• Recommendation and personalization algorithms
• Pricing optimization frameworks
• Campaign response modeling and attribution

These aren't tools in the same sense as short-video-factory. They are annotated, runnable blueprints. A data science team takes a tensor-house notebook, adapts it to their data schema and infrastructure, and ships it as part of a larger marketing analytics or automation platform.

The project's 1,440 stars and continued maintenance signal that teams are actively using it as a starting point — not just reading it for educational purposes.

Who Benefits from tensor-house

• Enterprise marketing engineering teams building custom ML pipelines for personalization or budget allocation
• Data scientists who need a validated starting architecture before writing production code
• Researchers and students studying applied ML in business contexts
• Marketing technologists at companies with proprietary customer data that makes off-the-shelf tools a poor fit

Limitations to Acknowledge

tensor-house requires genuine ML engineering fluency. Setting up the environment, understanding the model assumptions, and adapting notebooks to real data is not a weekend project for a marketer without a technical background. The repository provides reference implementations, not turnkey solutions. Time-to-value is measured in weeks or months, not hours.

Head-to-Head: Key Dimensions

Speed to First Output

Winner: short-video-factory. A desktop install and a product image can produce a finished short video in a session. tensor-house requires environment setup, data preparation, and model adaptation before it generates anything useful for your specific business.

Depth of Marketing Intelligence

Winner: tensor-house. short-video-factory automates creative production. tensor-house enables the analytical layer that decides which creative to produce, for whom, at what price, and through which channel. They operate at different layers of the marketing stack.

Scalability

Both scale, but differently. short-video-factory scales through batch processing — more assets in, more videos out. tensor-house scales through infrastructure — once deployed, ML models score millions of customers or transactions continuously. The relevant scale is determined by your problem.

Team Requirements

Maintenance Burden

short-video-factory is a desktop application — the maintenance is largely handled by the project maintainers through releases. tensor-house notebooks require ongoing ownership: your team needs to retrain models as data drifts, update dependencies, and adapt implementations as your marketing stack evolves.

Open-Source Context: Affitor and BrianRWagner Skills

Two adjacent projects also surfaced in this research cycle, both worth a brief mention for context.

Affitor/affiliate-skills (403 stars, updated 2026-05-18) offers "50 AI agent skills for affiliate marketing" including content research, post writing, infographic generation, and landing page deployment — described as a "full flywheel with social intelligence." It's compatible with Claude Code, ChatGPT, Gemini, Cursor, and others.

BrianRWagner/ai-marketing-claude-code-skills (296 stars, updated 2026-05-17) describes itself as "marketing frameworks that AI actually executes," designed primarily for Claude Code.

These two are in a different category from short-video-factory and tensor-house — they're agent skill libraries rather than standalone tools — but they illustrate how the AI-for-marketing open-source ecosystem is maturing beyond single-purpose scripts toward composable, agentic workflows.

Verdict: Which Should You Use?

The honest answer is that short-video-factory and tensor-house don't compete — they sit at opposite ends of the marketing automation spectrum.

Choose short-video-factory if:

• Your bottleneck is content production volume, specifically short-form video
• You need output today, not in six weeks
• Your team doesn't include ML engineers
• You're in e-commerce, social media, or content marketing

Choose tensor-house if:

• You have a data science or ML engineering team
• You need custom models trained on your own customer or transaction data
• Your use case is marketing analytics, personalization, or budget optimization — not content creation
• You're building a proprietary marketing intelligence system that off-the-shelf SaaS tools can't serve

Use both if you're building a sophisticated marketing operation: tensor-house models can inform which products to feature and which segments to target, while short-video-factory can execute the creative production at scale once those decisions are made.

---

Both projects were updated on the same day this article was researched (2026-05-18), suggesting active and healthy maintenance communities. For the latest capabilities, check the respective GitHub repositories linked at the top of this article.