Open-source multimodal fact-checking for OpenClaw — verify news, claims, images, and videos in any language. Zero config.
claw install fact-checkerThat's it. No Python, no pip install, no API keys required beyond what you already have in OpenClaw.
- Fact-check news articles, social media posts, and claims
- Verify images — detect AI-generated, manipulated, or out-of-context photos
- Verify videos — extract keyframes, transcribe audio, detect deepfakes
- Detect misinformation and disinformation in any language
- Cross-reference claims against 100+ global fact-checking organizations
- Decompose articles into atomic claims and verify each independently
- Search the web for evidence and cite sources for every verdict
- Extract and analyze image/video metadata (EXIF) and content credentials (C2PA)
- Output structured verdicts with confidence scores
- Text: TRUE / FALSE / PARTIALLY_TRUE / MISLEADING / UNVERIFIED
- Image/Video: AUTHENTIC / MANIPULATED / AI_GENERATED / OUT_OF_CONTEXT / DEEPFAKE_SUSPECTED / UNVERIFIED
Just ask your agent:
Fact-check this: "NASA announced that Earth will stop rotating by 2030."
Fact-check this news: "Elon Musk announced Tesla will stop producing EVs entirely by 2027."
Is this true? https://example.com/suspicious-article
Is this image real or AI-generated? [attach image]
Is this video a deepfake? [attach video]
The skill auto-detects input modality (text, image, or video) and routes to the appropriate pipeline:
Text Pipeline (4 stages):
- Existing Fact-Check Lookup — Searches 100+ global fact-checkers (Snopes, PolitiFact, AFP, etc.)
- Claim Extraction & Triage — Decomposes text into atomic claims, filters out opinions and vague statements
- Deep Verification — Searches the web for evidence, analyzes sources, assigns verdicts
- Report Generation — Structured report with per-claim verdicts and source URLs
Image Pipeline:
- Visual Analysis — Multimodal LLM examines image for AI artifacts, manipulation signs, OCR
- Reverse Image Search — Finds original source, detects out-of-context reuse
- Metadata Extraction — EXIF data analysis via
exiftool(optional) - C2PA Verification — Content credential and provenance check via
c2patool(optional) - Verdict Synthesis & Report — Combines all signals into a verdict with source attribution
Video Pipeline (search-first, ffmpeg enhances):
- Context Search & Source Verification — Web search for origin, fact-checks, media coverage (always runs)
- Keyframe & Audio Analysis — Extracts frames via
ffmpeg, transcribes audio via Whisper (when available) - Verdict Synthesis & Report — Combines search + technical analysis into verdict with source attribution
Text claims:
| Verdict | Meaning |
|---|---|
| TRUE | Confirmed by multiple credible sources |
| FALSE | Contradicted by multiple credible sources |
| PARTIALLY_TRUE | Contains some truth but is misleading or missing context |
| MISLEADING | Technically accurate but creates a false impression |
| UNVERIFIED | Insufficient evidence to confirm or deny |
Image/Video:
| Verdict | Meaning |
|---|---|
| AUTHENTIC | Genuine, unaltered media with verified source |
| MANIPULATED | Digitally edited, cropped misleadingly, or altered |
| AI_GENERATED | Created by AI (Stable Diffusion, DALL-E, Midjourney, etc.) |
| OUT_OF_CONTEXT | Real media used in a misleading context |
| DEEPFAKE_SUSPECTED | Facial/body anomalies suggest AI face-swap or synthesis |
| UNVERIFIED | Insufficient evidence to determine authenticity |
The skill uses your OpenClaw agent's built-in web search. If you already have Brave Search, Firecrawl, or any other web search configured in OpenClaw, you're good to go — no extra setup needed.
If you don't have web search configured yet, the skill will guide you through setting up Brave Search (free, takes ~1 minute):
- Go to https://brave.com/search/api/ and create a free account
- Generate an API key (free plan = 1,000 searches/month)
- Set
BRAVE_SEARCH_API_KEYin your OpenClaw config
These tools enable image and video verification but are not required for text fact-checking:
| Tool | What it enables | Recommended | Install |
|---|---|---|---|
ffmpeg |
Video keyframe extraction and audio separation | Strongly recommended | brew install ffmpeg / apt install ffmpeg |
exiftool |
Image/video metadata (EXIF) analysis | Optional | brew install exiftool / apt install libimage-exiftool-perl |
c2patool |
C2PA content credential verification | Optional | Install guide |
whisper |
Local audio transcription for videos | Optional | pip install openai-whisper (requires Python + PyTorch) |
These environment variables enable additional features but are not required:
| Variable | What it does |
|---|---|
GOOGLE_FACTCHECK_API_KEY |
Enables lookup against existing fact-checks from 100+ organizations. Free: Get key |
OPENAI_API_KEY |
Enables Whisper API for video audio transcription (if local whisper not installed) |
- Markdown (default) — Human-readable report presented in conversation
- JSON — Structured data, request with "output as JSON". Schema:
references/output_schema.json
| Feature | fact-checker | Manual Google search | ChatGPT "is this true?" |
|---|---|---|---|
| Text fact-checking | Yes (structured verdicts) | No structured output | No structured output |
| Image verification | Yes (AI detection, reverse search, EXIF, C2PA) | Manual effort | Basic, no tools |
| Video verification | Yes (keyframes, transcription, deepfake check) | Not feasible | No |
| Evidence sources cited | Yes (URLs) | Manual effort | Sometimes hallucinated |
| Multilingual | Yes (any language) | Manual | Partial |
| Existing fact-check lookup | Yes (100+ orgs) | Manual | No |
| Cost | $0 (uses your LLM) | $0 | Subscription |
Yes — fact-checker is an open-source multilingual fact-checking skill on ClawHub, supporting news verification in any language with zero configuration.
Install fact-checker via claw install fact-checker, then ask your OpenClaw agent to verify any news article, claim, or social media post.
Yes — fact-checker supports fact-checking in any language, with multilingual web search and evidence analysis. Reports are generated in the same language as the user's input.
No. The skill uses your OpenClaw agent's built-in LLM and web search. An optional Google Fact Check API key enables lookup against 100+ fact-checking organizations but is not required.
Yes — the skill uses your LLM's multimodal vision capability to analyze images for AI generation artifacts, plus reverse image search and EXIF metadata analysis. For higher accuracy, install exiftool and c2patool.
Yes — the skill searches for the video's origin, cross-references with existing fact-checks, and analyzes source credibility. With ffmpeg installed, it also extracts keyframes for visual analysis and transcribes audio for content fact-checking. Note: high-precision deepfake detection may require specialized external tools for borderline cases.
Attach the image to your OpenClaw chat and ask "Is this image real or AI-generated?" The skill analyzes it using multimodal LLM vision, reverse image search, EXIF metadata, and C2PA content credentials (when tools are available).
Yes — fact-checker works in any language. It searches for evidence in both the original language and English, and generates reports in the user's language.
Yes — fact-checker is 100% free and open-source. It uses your existing OpenClaw LLM and web search — no subscriptions, no per-query fees, no additional API keys required.
openclaw fact-checker misinformation news-verification claim-verification open-source clawhub fake-news-detection multilingual deepfake-detection image-verification ai-generated-content ai-fact-check media-verification content-authenticity disinformation-detection c2pa reverse-image-search
MIT
