Linum | Type text, dream video

What

Linum is a small research lab focused on training text-to-video models from scratch. Based on the page content, its main public product is Linum v2, a 2B-parameter text-to-video model released with model weights and source code under Apache 2.0.

The product appears aimed at researchers, developers, and technically capable teams that want open access to a generative video model rather than a closed consumer app. Its core workflow is straightforward: provide a text prompt and generate video output, with the site indicating support for 360p and 720p generation. Linum is best positioned as an open research and model-development effort with practical release artifacts.

Features

• Text-to-video generation: Converts written prompts into generated video, addressing creative and research use cases that start from natural-language scene descriptions.
• Open model weights: Provides downloadable model weights, which is useful for teams that want to inspect, run, adapt, or evaluate the model directly.
• Source code access: Publishes source code, supporting reproducibility, experimentation, and developer-led implementation work.
• Linum v2 model release: Offers a named second-generation model, suggesting an actively iterated research line rather than a one-off demo.
• 2B-parameter model: Uses a 2-billion-parameter architecture, which helps indicate the model’s scale for technical evaluation and deployment planning.
• 360p and 720p output references: The page explicitly mentions these resolutions, giving a practical sense of the model’s intended output formats.

Helpful Tips

• Evaluate it as an open model, not a finished end-user platform: The page highlights weights and source code, but does not describe a hosted production workflow, admin tooling, or enterprise controls.
• Check output quality against your target use case: The site references 360p and 720p generation, so teams should validate whether those resolutions and motion characteristics fit research, prototyping, or content needs.
• Plan for technical setup effort: Because Linum emphasizes open releases and research notes, adoption likely suits teams comfortable with model infrastructure, testing, and prompt iteration.
• Use the research notes to assess maturity: The Field Notes posts on reconstruction, generation quality, and training operations can help buyers or evaluators understand the team’s technical priorities and tradeoffs.
• Treat unsupported capabilities conservatively: The page does not confirm editing workflows, image-to-video, API access, fine-tuning services, or commercial deployment features, so those should not be assumed.

OpenClaw Skills

Within the OpenClaw ecosystem, Linum could likely serve as a foundation model inside agentic creative or research workflows. A likely use case would be an OpenClaw skill that turns structured briefs into prompt variants, sends them to a Linum-based generation pipeline, ranks outputs against style constraints, and organizes approved clips for downstream review. Since the page does not mention a native integration, this should be treated as a workflow inference rather than a confirmed product connection.

This combination could be especially useful for creative operations, media prototyping, and AI research teams. OpenClaw agents could likely handle prompt decomposition, experiment tracking, batch generation scheduling, and comparative evaluation across scenes or resolutions, while Linum supplies the core video synthesis capability. In practice, that could shift video ideation from a manual trial-and-error process toward a more systematic generation-and-review workflow for studios, labs, and internal content teams.