FireRed OpenStoryline
Direct media, scripts, voiceovers, and edits in natural language, then preserve the workflow as a reusable video style.
The README discloses API-key configuration, online media search and downloads, optional third-party AIGC services, major upstream projects, demo-asset ownership, and Apache-2.0 attribution. Deductions apply because the supplied files do not show least-privilege controls, per-action confirmation gates, a complete data-flow map, concrete secret-storage or redaction measures, vulnerability management, a bounded external-effects inventory, or rollback and recovery controls; transparency and human involvement are largely asserted.
Installation, startup commands, skill names, and the documented project structure are broadly self-consistent. Most Python dependencies are pinned, multiple operating systems and Docker are covered, and fallback skill-install commands are provided. Deductions apply because resource archives, models, FFmpeg, external APIs, and generation services create substantial availability dependencies, tomli is unpinned, and the evidence contains no runtime error taxonomy, actionable failure diagnostics, or degradation strategy.
The intended audiences and scenarios are unusually clear, spanning beginners, creative users, reusable styles, and batch production, with CLI, Web, MCP, Docker, OpenClaw, Claude Code, and experimental agent paths. Deductions apply because boundaries are expressed mainly through TODO items and a few quality or cost warnings, while natural-language trigger conditions, ambiguity handling, skill conflicts, and platform-specific capability differences are not precisely specified.
The README has strong information architecture and detailed cross-platform installation and startup notes, while the full LICENSE matches the stated Apache-2.0 metadata. Deductions apply because linked tutorials and the FAQ are absent from the supplied evidence, naming varies slightly among OpenStoryline, FireRed-OpenStoryline, and skill identifiers, NEWS is not a complete changelog, the Docker tag is not mapped to the reviewed revision, and maintainers, support expectations, reporting channels, and release ownership are unclear.
Media retrieval, planning, scripting, narration, music, detailed editing, and reusable Style Skills describe a useful end-to-end workflow with several showcased output categories and notes about resolution and asset modes. Deductions apply because effectiveness is supported mainly by descriptions and demos rather than static quality benchmarks, representative timing, resource requirements, or comparative cost evidence; only AI-transition cost and unpredictability receive a concrete caution.
Pinned dependencies, commands, directory descriptions, a Docker tag, dated NEWS entries, TODO items, and explicit warnings make some claims traceable, and the LICENSE corroborates the licensing claim. Deductions apply because only the README, LICENSE, and requirements are supplied: implementation, configuration, skill definitions, tests, and linked documentation are unavailable, so claims such as enterprise-grade reliability, immediate results, and precise style replication cannot be cross-checked. Cost, quality, experimental compatibility, and planned capabilities are nevertheless separated reasonably well from established features.
- Before providing private footage or production credentials, verify config.toml permissions, log redaction, key lifecycle, and exactly which services receive media or prompts.
- Online media retrieval, third-party AIGC, models, and resource archives introduce network, copyright, supply-chain, and billing risks; inspect download scripts, integrity checks, and asset licenses in an isolated environment.
- The evidence does not establish pre-edit confirmation, operation auditing, or project-level undo; use read-only copies of source media and avoid overwriting originals until outputs are validated.
- Although most dependencies are pinned, no lockfile, hashes, vulnerability scanning, or update policy are evidenced, and tomli is unpinned.
- README demos and reliability claims were not executed or independently tested in this static assessment and should not be treated as guarantees of performance, quality, or determinism.
What does this agent do, and when should you use it?
FireRed OpenStoryline is a self-hosted AI video-editing agent that translates conversational direction into concrete editing operations. Its core application under src/open_storyline contains an MCP server, video-processing nodes, Skills, storage, prompt templates, and agent construction code. It can find and download media, segment and understand footage, draft scripts and narration, and refine clip order, captions, music, and visual details through natural-language requests. Workflows are available through cli.py, a FastAPI-based web service, or MCP, and they produce edited videos rather than merely written editing instructions. A completed editing workflow can also be archived as a reusable Skill, although setup requires local resources, an API key in config.toml, and external services for some functions.
A user supplies a theme, directions, and source media, while the system can also search for and download matching images and video clips online. Processing nodes under src/open_storyline/nodes segment and analyze footage; the system combines the theme, visual understanding, and emotion recognition to construct a storyline, narration, and scripts that can imitate the tone, rhythm, and sentence structure of reference copy. It recommends background music, voiceovers, and fonts, supports personal playlist imports and beat synchronization, and executes conversational requests to cut, replace, or resequence clips, revise scripts, and change color, font, stroke, or position. open_storyline.mcp.server exposes the MCP service, cli.py supplies a command-line conversation, and agent_fastapi:app supplies the web interface. Users can archive the resulting workflow as a custom Skill and apply it to replacement media. The repository also documents an ASR rough-cut Skill that removes filler words, disfluencies, and repeated sentences with timestamp-aligned segmentation, plus AI transitions generated from adjacent boundary frames and a text description through third-party AIGC video services.
- A beginner or independent creator who wants to assemble a travel vlog, unboxing video, or year-in-review by describing edits conversationally.
- A content team producing a recurring series that wants to save one complete editing workflow as a Skill and reuse it with new footage.
- A review, recommendation, or casual-vlog creator who needs scripts modeled on the voice, pacing, and sentence patterns of reference copy.
- An editor cleaning interviews, explainers, or other speech-led footage who wants timestamp-aligned removal of fillers, disfluencies, and repeated sentences.
- A team seeking a coherent audiovisual style that wants content-aware BGM, voiceover, and font recommendations with beat syncing.
- A developer incorporating video editing into a broader agent workflow through the MCP server, OpenClaw Skills, built-in Claude Code Skills, or experimental Codex Skills.
What are this agent's strengths and limitations?
- Covers a concrete pipeline from media discovery and footage understanding through scripting, narration, editing, music, voiceover, and font selection.
- Supports iterative natural-language control over clip cuts, replacements, ordering, script text, and detailed caption styling with immediate results.
- Archives an entire editing workflow as a custom Skill, enabling style reuse across replacement footage and batch-oriented production.
- Offers CLI, FastAPI web, MCP, and Docker delivery modes, along with built-in Claude Code Skills and documented OpenClaw support.
- Includes specialized speech rough cutting with timestamp alignment and AI transition generation based on the boundary frames of neighboring clips.
- Setup involves a Python environment, model and resource downloads, dependency installation, and API-key configuration; Windows requires manual resource placement.
- AI transitions use third-party AIGC video generation services, carry relatively high costs, and can produce unpredictable results depending on footage, prompts, and model performance.
- Licensing restrictions limit the bundled open-source fonts and music to basic effects; stronger production results require a custom asset library.
- Voice cloning, broader transitions and effects, image and video generation or editing, GPU rendering, and highlight selection remain TODO items.
- Codex and other compatible-agent Skills are described as experimental, so their maturity is not equivalent to the repository's built-in Claude Code workflow.
- Online media search, playlist-related features, configured APIs, and transition generation introduce network, credential, and external-service dependencies.
How do you install or deploy this agent?
Python 3.11 or later is recommended, with Conda used for environment management:
git clone https://github.com/FireRedTeam/FireRed-OpenStoryline.git
cd FireRed-OpenStoryline
conda create -n storyline python=3.11
conda activate storylineOn Linux or macOS, the automated setup is:
sh build_env.shFor manual installation, install wget and run:
chmod +x download.sh
./download.sh
pip install -r requirements.txtInstall torchaudio in the same environment if storyline.local_asr will be used. On Windows, create resource in the project root, extract models.zip into .storyline, extract resource.zip into resource, and run pip install -r requirements.txt. Before launch, add the required API key configuration to config.toml. A Docker image, openstoryline/openstoryline:v1.0.1, is also documented; its run command mounts config.toml, outputs, and run.sh and publishes port 7860.
How do you use this agent?
Configure the required API key in config.toml first. Start the MCP server on macOS or Linux with:
PYTHONPATH=src python -m open_storyline.mcp.serverOn Windows use:
$env:PYTHONPATH="src"; python -m open_storyline.mcp.serverThen launch the CLI:
python cli.pyOr run the web service:
uvicorn agent_fastapi:app --host 127.0.0.1 --port 8005Provide the theme, editing instructions, and media through the conversation interface, then issue further natural-language requests to cut, swap, resequence, rewrite, or adjust visual styling. From the repository root, Claude Code can invoke /openstoryline-install and /openstoryline-use. The documented experimental Codex path uses npx skills add FireRedTeam/FireRed-OpenStoryline --skill openstoryline-install --agent codex, followed by the equivalent command for openstoryline-use.