Let Claude Help You Make News and Documentary Shorts: 8 Styles, from a Single Short Film to Your Own Recurring Series

Hand over reports, archives, or game footage to Claude to produce a news or documentary short film; once you find a visual style that works for you, save it and reuse it with new materials next time.

Lemomo has updated Lemo-Opuscar, an open-source code-based filmmaking tool for Claude, introducing two substantial changes: first, eight new "Editing and Documentary" styles designed specifically for cutting real-world footage, expanding the ready-to-use official style library to 59 styles; second, the official launch of the "Customize Your Own Style" and "Start from Scratch" workflows, allowing creators to adapt official presets or custom rules into reusable personal styles.

Eight New Styles: How to Cut Real Footage into Film

In the eight newly added "Editing and Documentary" styles, the core change is that base visuals no longer rely solely on code-generated graphics from scratch; instead, they use real-world documents, video footage, photographs, and verified data as their foundation.

When releasing these eight styles, alongside the primary demo reels produced with Claude Opus 5.5, the author also released test demos using Sonnet 5.5 High. The two sets of demos feature different subject matter and runtimes; while they showcase what each can achieve, they do not constitute a controlled model-performance benchmark.

1. Caption News

Using authentic live-action footage, on-site photographs, and public documents as base layers, facts remain fully conveyed even when played muted on social platforms. When handling official PDF reports (such as agricultural data bulletins), the code renders them as realistic white paper laid out on a dark desk, sweeping a highlighter stroke across the original text position and extracting magnified figures alongside.

Oversized Chinese captions serve as visual focal points, with core keywords set against honey-yellow color blocks; by frame 4, giant key metrics bounce into view and roll to a stop; whenever data and documents appear, a Source tag slides in synchronously at the upper-left of the screen.

Suggested applications: Feed breaking news, data investigations, and fact-based reports with narrative turns.

2. 1920s Newsreel

Sourced from authentic historical archival footage, vintage newspaper layouts, and vintage gramophone records, avoiding artificial aging effects or generative illustrations. Scans and vintage reels undergo percentile contrast stretching and grayscale mapping, standardized into a 1.37:1 rounded-corner aspect ratio with superimposed gate weave, subtle projector flicker, and vertical scratches. Searchlights and a spinning globe form a retro opening, headline newspapers spin forward toward the camera, the announcer speaks at rapid-fire speed (simulating early 1930s optical sound-on-film track audio), energetic jazz cuts to the musical beat, stamped location cards expand, and hand-drawn pencil circles freeze on key moments.

Suggested applications: Historical anecdotes, evolution of vintage industrial artifacts, and retro brand shorts.

3. Live Countdown Split-screen

The core constraint is that screen time strictly equals real-world event time, with zero jump cuts. If multi-angle video feeds capture the same live moment, they must synchronize to real physical event time; photographs and schedule documents are inserted as reference cells, while asynchronous video clips bear explicit unsynchronized corner tags (such as "Same window · Not frame-synced") to prevent misunderstanding. A central clock governs the entire sequence, ticking once every full second; as zero hour approaches, the split-screen divisions multiply (displaying control rooms, procedural checklists, and status logs); the instant zero is reached, the split screens resolve into a full-screen close-up, and the clock background inverts to an orange background with black lettering for "T+ elapsed time."

Suggested applications: Rocket launches, match-ending final whistles, major retail product launches, and lottery drawing moments.

4. Time-lapse Symphony

Combining real-world city fixed-camera time-lapses, overhead aerial shots, and historical archival footage. Multi-frame smoothing algorithms convert fast-moving traffic and pedestrians into flowing light trails; portraits avoid excessive close-ups, treating crowds as kinematic movement data. The entire film features zero voiceover narration, driven by a steadily building orchestral symphony; fixed-angle long time-lapses recur throughout as an environmental chronometer; climaxes slam to a halt on orchestral downbeats into an instantaneous freeze frame, or cut straight to black, leaving only reverb trails.

Suggested applications: City promotional showcases, large-scale civil engineering documentaries, and regional landscape transformations.

5. Visual Investigation

Built on an investigative workbench logic, five independent categories of evidence (live video footage, eyewitness photos, investigative reports/telegrams, original audio recordings, and survey maps) are aligned and cross-verified along a unified timeline. Viewing cones on floor plans are tagged as inferred when deduced; verified official findings remain strictly distinguished from unresolved questions. A stark, dark workbench aesthetic. Key milestones freeze frame, highlighting details with red circles and localized magnifications; 2D floor plans track eyewitness positions in real time; known facts and open questions sit side by side in parallel columns.

Suggested applications: Incident reconstructions, debunking reports, accident forensics, and controversial ruling verifications.

6. Constructivist Montage

Historical archival photographs, engineering documentation, and period propaganda pamphlets are transformed into a tri-color print texture of red, black, and paper-white. Cutout portraits intentionally retain raw white paper halos, layered together to construct dynamic visual geometry. Typography aligns along diagonal -18° steel-beam color bands; hard cuts fire at a mechanical 120 BPM punch-press tempo, juxtaposing colliding images to generate symbolic metaphors; worker portraits and numeric counters climb upward tier by tier along structural beams.

Suggested applications: Industrial history, milestone engineering completions, portraits of the working class, and period commentary.

7. Sports Telestrator

Tactical breakdowns rely on game tape freezes and on-screen telestration: timing measured frame by frame, positions and spacing estimated against field markings and pitch perspective; unverified metrics like ball velocity or running distance are never asserted if not directly observable. Guidelines require cropping original broadcast watermarks and logos, and external footage still requires proper usage rights. Studio replay structure: regular-speed goal playback ➔ tape-noise rewind ➔ keyframe freeze flash ➔ fluorescent player circles and passing vectors drawn in true grass perspective ➔ defensive gaps highlighted ➔ camera rotation flattening the pitch into an overhead tactical board.

Suggested applications: Decisive play breakdowns, tactical analysis, and competitive live-action clip dissections.

8. Broadcast News Package

A complete television news graphic suite generated entirely through code. Backgrounds permit continuous schematic loops or dynamic textures; reporting subjects appear as typographic layouts to avoid unauthorized commercial enterprise logos; rolling tickers are eliminated across the film to ensure information remains stationary and legible. Opening and closing channel idents, rundown line-ups, persistent lower-third news straps, full-screen data chart comparisons, and map drill-downs.

Suggested applications: Weekly tech roundups, industry breaking briefs, periodic corporate briefings, and recurring series.

How Claude Turns Assets into Video

In Lemo-Opuscar, a STYLE.md prompt is not a rigid video template locking down specific pixels or preset visuals, but a specification describing rendering techniques, color palettes, camera work, motion, and acoustic texture. Upon reading it, Claude writes frame-by-frame rendering code to organize the audiovisual experience. The author chose to describe desired effects through prompts rather than hardcoding static parameters in order to inspire creators and broaden their thinking when exploring the library, while avoiding rigid constraints on the model so its inherent capabilities can shine and styles can evolve alongside model advancements; actual output still relies on real styleframes and short clips for preview and iteration.

Code functions here as the control switcher, effects unit, and packaging editor: it schedules assets against the timeline, annotates live footage with lines and highlights, aligns timing, and synthesizes charts with audio. Turning raw assets into finished video follows a concrete execution pipeline: creators prepare the necessary footage clips, Claude generates a web page embedding the rendering logic, and schedules video frames, text, and dynamic annotations according to time parameter t; a headless browser then executes sequential window.render(t) calls across specified frame numbers and frame rates to capture frame-by-frame screenshots; finally, ffmpeg encodes the image sequence into video and multiplexes the audio.

Modifying a Style: What Exactly Are You Changing?

Beyond using pre-built official styles, the most critical mechanism in this update is enabling creators to build their own private style assets. If you simply want to alter a script, change voiceover narration, or switch actors, these are single-film narrative adjustments that can be handled directly in your prompt without creating a new style; building a custom style is only necessary when you need to standardize a long-term brand palette, specific typography, editing rhythm, or signature visual grammar.

The system provides two starting paths:

1. Adapting an existing library style: Pick one from the 59 available styles as a base and fine-tune only the elements you care about;
2. Starting from scratch: Describe your desired visual aesthetic in natural language, or provide Claude with reference images or video. Claude will automatically match the closest official style as a technique base, explaining the rationale and seeking your consent.

Regardless of the path chosen, style planning begins by working backward from practical utility: first determine the subject matter, the target audience, the publishing platform (for vertical video platforms, this requires matching 1080×1920 or 1080×1440 aspect ratios), and whether fixed brand colors, logos, fonts, or character designs are required; once the purpose is clear, select the specific modules to alter. Adjustments must correspond to intended utility, and adapting a library style caps adjustments at five modifications per cycle (supplementary Chinese fonts required for language localization do not count toward this limit of five) to prevent overcomplication. When planning concludes, Claude issues a proposal summary outlining what each item was before, what it becomes after, and the associated trade-offs, writing to draft only after explicit creator confirmation.

Changing colors, swapping materials, and altering core rendering techniques carry distinctly different impacts:

Reviewing a Still Styleframe First, Then Verifying Motion with a Short Demo

When establishing style specifications, the system validates the result in two progressive stages:

1. Code-rendered styleframe: Code executes directly to render a single static frame. If modifying an official style, a side-by-side comparison image (styleframe-vs-base.jpg) is generated, with the original on the left and the customized version on the right. Full video rendering is skipped at this stage, focusing first on verifying layout spacing, colors, and material textures;
2. 15–30 second short demo: Once the styleframe is approved, Claude generates a short video clip. Its primary role is not complex storytelling, but verifying that the style holds together dynamically—ensuring every adjustment recorded in the change log (CHANGES.md) appears at least once, showcasing signature camera moves, score, and sound effects (uncluttered by voiceover narration or subtitles by default), and timestamping the exact second each modification occurs.

How Swiss Typography Became a Tangible Physical Grid

In official examples, the author demonstrated transforming classic "Swiss Motion Graphics" into a "Material Grid" style.

These two representative frames illustrate the visual direction before and after adaptation: one is a flat poster, while the other is a three-dimensional arrangement of tangible physical objects.

Throughout this transformation, the author retained Swiss typography's core disciplined grid, clean sans-serif fonts, and beat-synced cuts, but overhauled the rendering materials and camera grammar: solid flat color blocks were replaced with three-dimensional physical textures like clay, plastic, felt, glass, and chrome; camera movement shifted from 2D planar zooms to "looking like a flat poster from top-down, but revealing three-dimensional physical depth the moment the camera tilts."

The following 31-second short demo verifies the dynamic viability of this style:

styles/
  caption-news/
  swiss-motion/
  ... (59 official styles)

README.md (Global index)
<slug>/
  style.json / STYLE.md
  CHANGES.md / DEMO.md
  demo/ (Standalone build source and build.sh)

Saving and Reusing Styles Over the Long Term

After generating the verification short, the design does not automatically become a global preset. Until the creator explicitly confirms saving it as a personal style, all files remain confined to the current project directory; you can request further revision rounds at any time, or choose not to save it as a personal style.

Once saving is confirmed, the system archives it into the local ~/lemo-my-styles/<slug>/ directory and appends an entry to the index in ~/lemo-my-styles/README.md, enabling retrieval by name in future sessions. The archive directory structure mirrors official styles:

• STYLE.md: Records fixed style rules independent of subject matter, including rendering techniques, color systems, camera grammar, and sound textures;
• CHANGES.md: Fully documents every specific modification decision and rationale relative to the base style;
• DEMO.md, demo/ source code, and sample footage: Includes a one-click build script build.sh, code, required micro-assets, licensing notes in CREDITS, and generated test clips and styleframes for inspecting technical implementations and rebuilding tests at any time.

Once saved, you can invoke the style directly by name in new projects (for example, "make a short film using my Material Grid style"). When producing new videos, the code reuses only the established visual specifications and implementation techniques, never copying the specific plot, storyboard, or timeline of the test demo. Furthermore, personal style directories remain separate from the official library; official library updates or plugin reinstalls will never overwrite personal assets. When upstream bug fixes are applied to the base style, Claude asks at launch whether you wish to merge them, leaving the decision to you rather than modifying personal versions autonomously.

Putting Your AI to Work on Filmmaking

If you already use coding AI tools like Claude Code that can inspect local files and execute commands, you can provide the repository URL, your source materials, and your production objectives together, allowing it to check the environment against project documentation, install or update tools using official methods, and complete production. First clarify who your audience is and what single takeaway they should remember.

For example, to turn your reports and footage into a 45-second short, you can send the following prompt, written for this guide. Replace the bracketed text with your actual asset paths. Note that this example requests Chinese output as written; you can adapt the target language for your own audience:

Read the installation and usage instructions at https://github.com/lemomo-ai/lemo-opuscar, check the environment, and install or update it using the official method. Tell me when I need to sign in or restart the session.

Use Caption News style to make a 45-second Chinese film from the reports and footage at [my material folder or file path]. First list the files you actually read, the proposed figures, statistical definitions, and sources, then propose a storyline. If material is missing, explain what is needed; do not invent figures. Show me a storyboard and wait for confirmation before production. Display sources beside key figures. Deliver the MP4, project path, and material source notes.

From this starting point, you can extend these workflows into your recurring content operations.

Turning Weekly Materials into a Recurring Series

If you interpret industry reports weekly, you can have AI first identify the most noteworthy shifts in the current issue, cross-check numeric definitions and sources, and then produce a short along a consistent visual style. Next week, swap in new materials while preserving the established color palette, typography, and editing cadence, gradually building a recognizable recurring series.

The same approach applies to highlight recaps of interview transcripts or tactical breakdowns of match recordings. First define what question the content answers, then select the appropriate visual language: identifying what exists in the data and what conclusions it supports requires collaborative review between you and the AI.

Testing Multiple Narrative Approaches on the Same Material

Having a single report does not limit you to a single style of video. You can prompt the AI to propose two or three creative treatments first, outlining a synopsis and key shots for each, before committing to render the full video.

For example, the same source materials could be shaped using "Caption News" to isolate the single most critical change for sound-off mobile browsing; or using "Broadcast News Package" to incorporate background context, data charts, and specification cards for a comprehensive delivery. If the assets comprise multi-source evidence requiring alignment and verification, consider "Visual Investigation." Style selection dictates how information is structured; exploring narrative approaches beforehand helps clarify how the story is best told.

Saving Account Identity into a Reusable Style

Once you find a visual treatment that fits your channel identity, prompt the AI to organize it into a personal style. For instance, you might adapt "Swiss Motion Graphics" for a weekly tech bulletin by retaining its grid and rhythm while adopting your brand colors and localized fonts; or provide reference images, specify the compositions, materials, and camera feel you want to preserve, and ask the AI to suggest adaptation directions.

Here is another prompt written for this guide, for creating a branded series style:

I want a personal style for my weekly tech bulletin. Start from Swiss Motion Graphics, keep its grid and beat-based cuts, change the accent to brand navy (#0A2540), use Noto Sans SC for Chinese, and speed up motion by 15%. Show me the proposed changes and a base-versus-custom styleframe comparison. After approval, make a 15–30-second demo. Only when I am satisfied, save it as “科技周报深蓝” (tech-bulletin-navy) and tell me how to invoke it next time.

When tackling new topics later, you can instruct the AI directly: "produce this episode using my 科技周报深蓝 style." The visual rules and implementation techniques carry forward, while the story, camera shots, and assets are freshly organized around the new topic.

Your primary role is supplying assets, setting goals, choosing narrative approaches, and assessing whether styleframes and final films meet expectations; dependency installation, rendering code, and error troubleshooting can be delegated to the AI. The author estimates that producing a single short typically requires roughly 30–60 minutes of Agent execution and consumes a notable volume of model tokens, making it best to start with a focused, small-scale topic.

Studio Workbench and Anime Styles: Upcoming Previews

A Desktop Chat Filmmaking Workbench in Claude

• Lemo Movie Studio workbench: An internal MCP workbench based on the Claude desktop app is currently in development. In the future, creators will not need to toggle back and forth in the terminal; style selection, storyboard review, styleframe inspection, and final film delivery can be handled visually within the desktop chat interface. The author plans to offer free access upon release.

• Next batch of styles: Japanese anime aesthetics: The author announced that the next batch will focus on Japanese anime styles. Because these styles require integrated image generation, execution environment requirements will be more demanding than the current code-only rendering pipeline; specific dependencies will be disclosed upon release;
• Style library roadmap: Progressively expanding the official style library to 99 styles.

Code Use and Demo Footage: Understanding the Dual Licensing Boundaries

While the project codebase is open-sourced under the permissive MIT license, different asset tiers carry distinct legal boundaries during actual video production:

• Tool code and documentation: Licensed under the MIT License;
• Official demo videos and still frames by the author: Released under the Creative Commons Attribution 4.0 International license (CC BY 4.0). You may share, adapt, and use them commercially, provided you credit "LemoLab," link to the project repository (github.com/lemomo-ai/lemo-opuscar), and indicate if changes were made;
• Original videos created by readers using this tool: Owned entirely by you, with no attribution required to LemoLab;
• Third-party asset licensing: Public domain archives, Creative Commons footage, and scientific data referenced in demos retain their respective original licenses, detailed in each style's CREDITS file; creators are responsible for clearing rights for any external assets imported into their own films;
• Three voiceovers excluded from CC BY 4.0: The following three voiceover recordings used by the author in the demo clips are excluded from the CC BY 4.0 license and remain governed by their respective commercial provider terms, meaning they cannot be extracted or reused arbitrarily:
  1. System announcements in Game Show Flat (macOS native say voice);
  2. Voiceover narration in Paper-cut Lightbox (Microsoft Edge TTS);
  3. Spoken commentary in Broadcast News Package (Volcengine Doubao TTS). When producing commercial videos, readers may use Kokoro, the recommended offline text-to-speech engine, or supply their own licensed voiceover tracks.