Loading...
AI Watermark Detector: Find Hidden Characters in Any AI Text
What Does This Detector Do?
The AI Watermark Detector scans text from any AI language model—ChatGPT, Claude, Gemini, DeepSeek, Grok, LLaMA, Perplexity, and others—to identify hidden watermarks and invisible tracking characters. While these characters are imperceptible to human eyes, they can cause technical problems, interfere with text processing, and serve as tracking mechanisms.
This tool provides detailed analysis showing exactly what types of hidden characters are present, where they're located, and how many exist in your text. Whether you're debugging mysterious code failures, investigating copy-paste issues, or simply curious about what your AI model is embedding in its output, this detector reveals the invisible layer beneath your visible text.
Unlike watermark removers that clean text, this detector preserves everything while highlighting problems. It's perfect for understanding what's happening with AI-generated content before deciding whether to clean it, helping you make informed decisions about handling text from ChatGPT, Claude, Gemini, or any other AI platform.
Why Detect Watermarks Before Removing Them?
Detection before removal is valuable for several reasons. First, it helps you understand what types of watermarks different AI models use. You might discover that ChatGPT uses certain hidden characters while Claude uses different ones. This knowledge helps you identify which AI generated specific text, useful for tracking sources in mixed-content workflows.
Second, detection helps diagnose technical problems. When AI-generated code won't compile or URLs don't work, the detector pinpoints exactly where problematic invisible characters are located, making debugging much easier. You can see if the issue is a zero-width space in a variable name, a non-breaking space in a string literal, or some other specific hidden character.
Third, detection provides verification after cleaning. Run the detector before and after using a watermark remover to confirm that all hidden characters were successfully eliminated. This quality control step ensures your cleaned text is truly watermark-free and ready for use in production environments.
Types of Hidden Characters Detected
- Zero-width spaces (U+200B): Completely invisible spaces often used in watermarking schemes by various AI models.
- Zero-width non-joiners (U+200C): Invisible characters that prevent character joining in certain languages, sometimes used as markers.
- Zero-width joiners (U+200D): Invisible format characters that force character joining, occasionally embedded by AI systems.
- Word joiners (U+2060): Invisible characters that prevent line breaks at specific positions.
- Non-breaking spaces (U+00A0): Spaces that look normal but prevent line breaks and behave differently than regular spaces.
- Soft hyphens (U+00AD): Invisible hyphens that only become visible during line breaks, used in some text formatting.
- Byte order marks (U+FEFF): Unicode markers that can accidentally appear in AI-generated text.
- Various Unicode control characters: An extensive range of invisible formatting and control characters across the Unicode standard.
The detector identifies all these character types and provides detailed information about each one found, including its Unicode code point, frequency, and exact positions in your text.
Universal Compatibility Across AI Models
Different AI companies implement watermarking differently. OpenAI's ChatGPT might use specific patterns of zero-width characters, Anthropic's Claude could employ different invisible markers, and Google's Gemini may have yet another approach. The AI Watermark Detector doesn't assume any particular AI model's behavior—it comprehensively scans for all types of invisible characters regardless of which system embedded them.
This universal approach means one tool works for analyzing content from ChatGPT, Claude, Gemini, DeepSeek, Grok, LLaMA, Perplexity, and any other AI model. You don't need different detectors for different AI platforms. As new models emerge with new watermarking techniques, this detector automatically handles them because it looks for fundamental Unicode characteristics rather than model-specific patterns.
The detector is equally effective with text from AI chat interfaces, API responses, automated generation systems, or any other source. If hidden characters are present, regardless of origin or purpose, the detector will find them.
Step-by-Step Detection Instructions
Step 1: Copy the text you want to analyze from ChatGPT, Claude, Gemini, or any other AI model. The text can be any length from a single sentence to entire documents.
Step 2: Paste the text into the detector's input box. The tool accepts text from any source and handles all languages and character sets.
Step 3: Click the "Detect" or "Analyze" button. The tool instantly scans every character in your text, identifying all invisible and control characters.
Step 4: Review the detailed report showing what types of hidden characters were found, how many of each type, and where they're located. The detector may highlight problematic positions in your text or provide a character-by-character breakdown.
Step 5: Use this information to decide whether to clean the text with a watermark remover, manually edit specific problem areas, or keep the text as-is if the detected characters aren't causing issues.
All detection happens locally in your browser with no server uploads. Your AI-generated content remains completely private during analysis.
Common Use Cases for Watermark Detection
Debugging Code Issues
When AI-generated code from ChatGPT, Claude, or other models fails mysteriously, watermark detection pinpoints the problem. A zero-width space in a Python function name or a non-breaking space in a JavaScript string might be invisible to your eyes but causes compilation or runtime errors. The detector shows you exactly where these problematic characters are located, enabling quick fixes and successful debugging.
Quality Control for Content Publishing
Before publishing AI-assisted content to websites, blogs, or marketing materials, use the detector to verify there are no hidden watermarks that might cause formatting issues or interfere with CMS systems. Clean content displays reliably across all platforms and browsers, while watermarked content might have unexpected behavior. Detection as part of your pre-publication workflow ensures quality and consistency.
Understanding AI Model Behavior
Researchers and AI enthusiasts use the detector to understand how different models watermark their output. By analyzing text from ChatGPT, Claude, Gemini, and other platforms, you can learn about their watermarking strategies and compare approaches across models. This research helps the AI community understand transparency, tracking, and verification mechanisms in language models.
Data Integrity Verification
Before importing AI-generated data into databases or data processing pipelines, detect watermarks that might interfere with operations. Hidden characters can cause queries to fail, comparisons to produce wrong results, or indexing to malfunction. Detection identifies these risks before they cause problems in production systems, enabling you to clean data proactively.
Privacy and Security in Detection
Watermark detection requires analyzing the complete content of your text, making privacy crucial. The AI Watermark Detector processes everything locally in your web browser using JavaScript—no text is ever transmitted to servers, APIs, or external services. Whether you're analyzing confidential business content, proprietary code, personal information, or any other sensitive material from ChatGPT, Claude, Gemini, or other AI models, it stays on your device.
This local processing approach means you can safely detect watermarks in trade secrets, client information, unreleased products, confidential research, or any other private content without security risks. No logs are created, no data is stored, and the tool works offline once loaded. For regulated industries or security-conscious environments, local detection is essential for maintaining data protection compliance.
The detector never modifies your original text—it only analyzes and reports. Your source content remains untouched, giving you full control over what to do with the information discovered.
Interpreting Detection Results
The detector provides detailed information about discovered watermarks. A report showing "23 zero-width spaces found" tells you the text contains significant watermarking. Results like "No hidden characters detected" confirm the text is clean and free of invisible markers. Position information helps you locate specific problems when debugging technical issues.
Different counts suggest different situations. A few scattered hidden characters might be accidental artifacts rather than intentional watermarks. Large numbers of systematically placed invisible characters likely indicate deliberate watermarking by the AI model. Patterns in the positions—like hidden characters at regular intervals—suggest structured watermarking schemes.
Compare detection results across different AI models to understand their different approaches. ChatGPT-generated text might show different watermark patterns than Claude or Gemini, helping you identify sources of mixed content or understand which models use which techniques.
Combining Detection with Removal
The AI Watermark Detector works perfectly alongside the AI Watermark Remover. Use this workflow: First, detect watermarks to understand what's present. Second, use the remover to clean the text. Third, detect again to verify complete removal. This three-step process ensures thorough cleaning and quality control.
For critical applications like production code, database imports, or published content, the detect-clean-verify workflow provides confidence that no hidden characters remain. It transforms an invisible problem into a visible, solvable one, giving you full control over your AI-generated content from ChatGPT, Claude, Gemini, or any other source.
Some situations might call for selective cleaning—removing only certain types of watermarks while preserving others. Detection results help you make informed decisions about which characters to remove and which to keep, enabling sophisticated content processing strategies.
Technical Applications and Advanced Use
Software developers use watermark detection in automated testing pipelines to verify that AI-generated code or documentation is clean before integration. Quality assurance teams incorporate detection into content review workflows to ensure published materials are free of hidden artifacts. Data engineers validate AI-generated datasets before loading them into production systems.
For developers working with AI model APIs (OpenAI, Anthropic, Google, or others), watermark detection helps monitor API response quality and ensure consistent output. You can programmatically detect watermarks in API responses and take appropriate action—logging incidents, automatically cleaning text, or alerting quality control teams.
Researchers studying AI watermarking techniques use the detector as an analytical tool to document watermarking prevalence, study patterns across models, and evaluate detection resistance. This research contributes to understanding AI transparency and accountability in language model deployment.
Handling False Positives and Edge Cases
Some legitimate text naturally contains characters the detector identifies. For example, certain languages require zero-width joiners for proper character rendering, and some professional typography uses non-breaking spaces intentionally. The detector shows you what's present—you decide what's problematic versus legitimate formatting.
Context matters when interpreting results. A few non-breaking spaces in formatted text might be normal typography, while hundreds of zero-width spaces scattered through code are almost certainly watermarks or artifacts. Use your judgment about what detected characters represent based on your specific text and use case.
If you're unsure whether detected characters are legitimate or problematic, test by removing them and seeing if the text still works correctly. For code, remove detected characters and check if compilation succeeds. For prose, remove them and verify readability is preserved. Empirical testing clarifies whether detection results represent real problems or false positives.
The Future of AI Watermark Detection
As AI language models evolve, watermarking techniques will likely become more sophisticated. Future models might use more complex patterns, multi-layer watermarking, or novel encoding schemes. The AI Watermark Detector is designed to remain effective regardless of these advances by focusing on the fundamental Unicode characters that any text-based watermarking must use.
New AI models from emerging companies will bring new watermarking approaches. Whether it's GPT-5, Claude 4, Gemini Ultra 2.0, or completely new AI systems, this detector will work with their output because it examines the fundamental character-level structure of text rather than model-specific patterns. This future-proof design ensures long-term utility as the AI landscape evolves.
Regulatory frameworks around AI transparency might eventually require standardized watermarking across all models. Detection tools will become increasingly important for verifying compliance, understanding content provenance, and maintaining control over AI-assisted work. This detector positions you ahead of those trends, providing capabilities that will only grow more relevant over time.
Frequently Asked Questions
Does this detector work with all AI models including ChatGPT, Claude, and Gemini?
Yes, the AI Watermark Detector works universally with text from any AI language model including ChatGPT (all GPT-3.5, GPT-4, and GPT-4 Turbo versions), Claude (Claude 2, Claude 3 Opus, Sonnet, Haiku), Google Gemini (Gemini Pro, Ultra), DeepSeek, Grok, LLaMA, Perplexity, and any other current or future models. The detector doesn't rely on knowing specific watermarking techniques used by individual AI companies. Instead, it comprehensively scans for all types of invisible Unicode characters that could serve as watermarks, making it effective regardless of which model generated the text or what watermarking method was employed.
Is my text kept private when I use this detector?
Absolutely. The AI Watermark Detector processes all text entirely within your web browser using JavaScript. Whether you're analyzing ChatGPT conversations, Claude outputs, Gemini results, or content from any other AI model, your text never leaves your device. There are no server uploads, no API calls to external services, and no data transmission whatsoever. This client-side processing approach ensures complete privacy for sensitive business content, confidential code, personal information, proprietary data, or any other private material. The tool even functions offline once loaded, and creates no logs or records of any kind.
What types of hidden characters can this detector find?
The detector identifies all types of invisible and control characters that AI models might use for watermarking including zero-width spaces (U+200B), zero-width non-joiners (U+200C), zero-width joiners (U+200D), word joiners (U+2060), non-breaking spaces (U+00A0), soft hyphens (U+00AD), byte order marks (U+FEFF), and various other Unicode control characters across the full Unicode range. It provides detailed information about each type found, including the Unicode code point, frequency count, and location within your text. This comprehensive scanning ensures that no hidden watermark characters escape detection, regardless of which AI model embedded them or what watermarking technique was used.
Does the detector modify or clean my text?
No, the AI Watermark Detector only analyzes and reports—it never modifies your text in any way. Your original content from ChatGPT, Claude, Gemini, or other AI models remains completely unchanged during detection. The tool simply scans for hidden characters and presents information about what it finds. If you want to actually remove detected watermarks, use the AI Watermark Remover tool (a companion to this detector). The separation between detection and removal gives you full control—you can analyze first to understand what's present, then decide whether cleaning is necessary for your specific use case.
Why would I detect watermarks instead of just removing them?
Detection serves several important purposes before removal. First, it helps you understand what types of watermarks different AI models use, enabling you to identify content sources in mixed workflows. Second, detection pinpoints exactly where problematic characters are located, crucial for debugging code that won't compile or URLs that won't work. Third, it provides quality control—you can detect before removal and again after to verify complete cleaning. Fourth, detection helps you learn about AI watermarking practices, understand your AI tools better, and make informed decisions about when cleaning is necessary versus when watermarks aren't causing problems. Knowledge is power, and detection provides visibility into the invisible layer of your AI-generated text.
How can I tell if detected characters are legitimate formatting or problematic watermarks?
Context and quantity help distinguish legitimate use from problematic watermarking. A few non-breaking spaces in formatted text are likely intentional typography, while hundreds of zero-width spaces scattered through code are almost certainly watermarks or artifacts. Some languages legitimately require zero-width joiners for proper rendering—these are normal, not problematic. Look at where detected characters appear: in natural positions (like after punctuation) suggests legitimate use, while seemingly random placement suggests watermarking. When in doubt, test by removing detected characters using the watermark remover—if the text still works correctly, they weren't serving a legitimate purpose. For code, if removal allows successful compilation, the detected characters were problematic.
Can this help debug why my AI-generated code won't run?
Yes, absolutely. This is one of the most valuable use cases for watermark detection. When code from ChatGPT, Claude, or other AI models looks correct but fails to compile or run, invisible watermarks are often the culprit. The detector shows you exactly where hidden characters are located—you might discover a zero-width space in a function name, a non-breaking space in a string literal, or invisible characters breaking syntax. With precise location information, you can target these problem areas for manual editing or use a watermark remover to clean the entire code block. This visibility transforms mysterious failures into solvable problems, saving hours of frustrating debugging time.
Does detection work with text in languages other than English?
Yes, the AI Watermark Detector works with any language that AI models can generate including Spanish, Chinese, Arabic, Japanese, French, German, Russian, Hindi, Korean, and hundreds of others. The detector operates at the Unicode level, examining the fundamental character structure of text regardless of language. Whether your AI model generated content in one language or mixed multiple languages (common with advanced multilingual models like GPT-4, Claude 3, or Gemini), watermark detection works identically. International characters, accents, tone marks, and special symbols don't interfere with detection—the tool simply looks for invisible control characters among any visible content.
How is this different from AI detection tools like GPTZero or Originality.ai?
Those are completely different types of tools serving different purposes. AI content detectors like GPTZero and Originality.ai analyze writing patterns, vocabulary, and sentence structure to determine if text was likely AI-generated. They look at linguistic features, not character-level watermarks. The AI Watermark Detector examines the hidden character layer beneath visible text, identifying invisible Unicode markers that AI models might embed. It doesn't assess whether text is AI-generated—it shows you what hidden characters are present regardless of origin. Use AI content detectors to verify content authorship; use watermark detection to find technical problems and hidden tracking mechanisms in any text.
Can watermark detection identify which AI model generated specific text?
Potentially, yes. If different AI models use distinct watermarking patterns—for example, ChatGPT consistently using zero-width spaces while Claude uses non-breaking spaces—detection results can suggest the source. However, this isn't foolproof because watermarking implementations change over time, not all models watermark consistently, and patterns might overlap. Detection gives you clues about content provenance but shouldn't be considered definitive attribution. For mixed-content workflows where you're combining outputs from ChatGPT, Claude, Gemini, and other models, watermark detection patterns can help you understand which sections came from which sources, aiding in content management and source tracking.
What does it mean if the detector finds zero hidden characters?
A result showing zero hidden characters means your text is clean and free of invisible watermarks or control characters. This could mean several things: the AI model that generated it doesn't use character-based watermarking, the text has already been cleaned using a watermark remover, the content wasn't AI-generated at all, or the particular generation didn't include watermarks. Clean text is ideal for technical applications, ensuring no hidden characters will interfere with code compilation, URL functionality, database operations, or text processing. If you're running detection after cleaning as a verification step, zero results confirm successful watermark removal.
How often should I detect watermarks in my AI-assisted workflow?
The frequency depends on your use case. For critical applications like production code, detect every time before deploying AI-generated content. For content publishing, detect before final publication to catch any formatting issues. For casual use or drafting, detection might be unnecessary unless you encounter problems. Develop a workflow based on risk—high-stakes uses warrant consistent detection, while low-stakes uses can be more casual. Many professionals detect watermarks whenever they copy significant content from ChatGPT, Claude, Gemini, or other AI models, treating it as standard practice like saving files or running spell-check. Regular detection prevents surprises and ensures consistent content quality.
Can I use detection results for research or documentation?
Yes, researchers studying AI watermarking techniques, transparency, and accountability use watermark detection to document practices across different models. You can record what types of watermarks ChatGPT, Claude, Gemini, and other models use, how frequently they appear, and how patterns change over time or across model versions. This research contributes to understanding how AI companies implement content marking and what implications exist for users. If publishing research findings, remember that watermarking implementations may change, so document the specific date, model version, and conditions of your analysis. Privacy considerations apply if your sample text contains confidential information—use the local processing nature of this tool to maintain data security while researching.
Does detecting watermarks affect performance or slow down my workflow?
Detection is extremely fast, typically completing in milliseconds even for lengthy text. The tool scans character-by-character using efficient JavaScript algorithms optimized for modern browsers. Unless you're analyzing massive documents (hundreds of thousands of words), you won't notice any delay. For typical AI outputs from ChatGPT, Claude, or Gemini—ranging from paragraphs to several pages—detection is essentially instantaneous. The performance impact is negligible, making detection practical for frequent use in production workflows, automated pipelines, or interactive editing sessions where you want immediate feedback about hidden character presence.
Can the detector find watermarks in API responses from AI models?
Yes, the detector works with text from any source including API responses from OpenAI's GPT API, Anthropic's Claude API, Google's Gemini API, or other AI service APIs. When you programmatically generate text using AI APIs, watermarks might be present just as in web interface outputs. You can integrate watermark detection into your application's processing pipeline—receive API response, detect watermarks, decide whether to clean, and proceed with your application logic. For automated systems generating large volumes of AI content, detection helps monitor content quality and ensure consistency. You might discover that certain API parameters or models produce more watermarks than others, enabling optimization of your AI integration.
What should I do if I detect watermarks in critical business content?
First, assess whether the detected characters are causing actual problems. If they're not interfering with your workflow or final output, they might not require action. If problems exist or you want clean content, use the AI Watermark Remover to eliminate hidden characters. For business-critical content, implement a detect-clean-verify workflow: detect to understand what's present, clean using the remover, detect again to confirm complete removal. Document your process for audit trails and quality control. Consider making watermark detection and removal part of your standard operating procedures for handling AI-generated content from ChatGPT, Claude, Gemini, or other models in business contexts.
Can watermarks be used to track who generated specific content?
Theoretically, yes. Advanced watermarking systems could encode user IDs, session identifiers, timestamps, or other metadata using patterns of invisible characters. Whether ChatGPT, Claude, Gemini, or other models actually implement such tracking isn't publicly documented, but the technical capability exists. Watermark detection can reveal that hidden characters are present, though decoding any encoded information would require knowing the specific encoding scheme. For privacy-conscious users, detection provides visibility into hidden tracking mechanisms that might exist, enabling informed decisions about whether to remove watermarks before sharing content or storing it long-term.
How does detection help with database and data processing workflows?
Before importing AI-generated data into databases or processing pipelines, watermark detection identifies invisible characters that could corrupt operations. Hidden characters can cause database queries to fail, string comparisons to produce incorrect results, indexing to malfunction, or exports/imports to break. By detecting watermarks before data entry, you can proactively clean problematic content and prevent mysterious database behavior that's difficult to debug. For data pipelines processing large volumes of AI-generated content from ChatGPT, Claude, Gemini, or other models, automated detection can flag problematic batches for cleaning, ensuring data integrity and reliable processing throughout your systems.
Does the detector work with formatted text like markdown or HTML?
Yes, the detector analyzes the underlying text content regardless of formatting. If you paste markdown or HTML from AI-generated sources, detection scans the full content including formatting codes and finds any hidden characters present. This is valuable because watermarks might be embedded within markdown syntax, HTML tags, or plain text portions. For web content or formatted documents from ChatGPT, Claude, or Gemini, detection ensures both the visible content and formatting codes are free of problematic invisible characters. Just be aware that legitimate HTML entities or markdown syntax aren't flagged as watermarks—the detector specifically targets Unicode control and invisible characters, not visible formatting marks.
Can I batch analyze multiple pieces of AI-generated content?
While the detector analyzes one text block at a time, you can concatenate multiple pieces and detect them all at once. For example, paste ten different ChatGPT responses together (perhaps with visible separators like "===" between each), run detection, and see results for the entire block. This batch approach is faster than individual analysis. However, position information in results might be less useful for batch analysis since you need to map detected positions back to individual pieces. For truly large-scale automated analysis (hundreds or thousands of documents), you might want to implement your own detection system using similar Unicode scanning logic, as repeated manual detection would be impractical.
What happens to emojis and special characters during detection?
Emojis, mathematical symbols, international characters, and other visible Unicode elements are not flagged as watermarks because they're visible, not hidden. The detector specifically targets control characters and invisible markers—characters with no visual representation. Your content from ChatGPT, Claude, Gemini, or other models can include any visible symbols, and they won't be incorrectly identified as watermarks. The detector is smart enough to distinguish between invisible characters (potential watermarks) and visible characters (legitimate content), even for unusual or rare Unicode symbols. Only truly invisible markers trigger detection results.
How will this detector handle future AI watermarking techniques?
The AI Watermark Detector is designed to be future-proof by targeting all invisible Unicode characters rather than specific watermarking implementations. Any future watermarking technique based on hidden characters—whether from GPT-5, Claude 4, Gemini Ultra 2.0, or entirely new AI systems—will be detected because the tool examines fundamental Unicode properties. Even if AI companies develop more sophisticated or subtle watermarking schemes, as long as they use character-based approaches (which is nearly inevitable for text), this detector will find them. The only watermarking that might escape detection would be non-character-based schemes like statistical word patterns, but those can't be "detected" in the same way since they're inherent to language structure rather than hidden characters.
Can detection results help me choose which AI model to use?
Yes, watermark detection can inform AI model selection based on your needs. If you require absolutely clean output without hidden characters (for example, generating code or technical documentation), you might prefer models that detection shows produce less watermarking. If watermarks don't affect your use case, you can focus on other model qualities like output quality, speed, or cost. By testing output from ChatGPT, Claude, Gemini, DeepSeek, Grok, LLaMA, Perplexity, and others, you can compare their watermarking practices and make informed decisions about which models best fit your workflow. Some users build this analysis into their AI tool evaluation process alongside traditional metrics like accuracy and performance.
Does detection interfere with copy-paste functionality?
No, detection is read-only and doesn't modify clipboard contents or text handling. You can freely copy and paste text before, during, or after detection without any interference. The detector simply reads what you paste into it and reports findings—it never alters what's in your clipboard or how copy-paste works in your system. This non-invasive approach means you can integrate watermark detection seamlessly into your existing workflows with ChatGPT, Claude, Gemini, or other AI models without changing how you normally handle text. Detection is purely analytical, providing information without side effects on your content or tools.
What if I disagree with what the detector flags as hidden characters?
The detector objectively identifies all invisible Unicode control characters based on Unicode standards—it doesn't make subjective judgments about whether they're "good" or "bad." If you believe flagged characters serve legitimate purposes in your text (like zero-width joiners required for certain language rendering), that's a valid assessment. Detection simply provides visibility; you decide what's appropriate for your context. If certain character types are legitimate in your domain, you can ignore those findings and focus on others. The detector gives you comprehensive information, and you bring domain expertise to interpret it. This partnership between automated detection and human judgment produces the best results.
Other ChatGPT Tools
ChatGPT Humanizer
Humanize ChatGPT text to bypass AI detection
AI Humanizer (Universal)
Humanize text from any AI model to bypass detection
ChatGPT Watermark Remover
Remove AI detection markers and invisible characters from ChatGPT text
ChatGPT Image Watermark Remover
Remove C2PA metadata watermarks from ChatGPT generated images
ChatGPT Image Watermark Detector
Detect C2PA metadata watermarks in ChatGPT generated images
ChatGPT Watermark Detector
Detect hidden characters and watermarks in ChatGPT text
ChatGPT Space Remover
Clean up extra spaces and blank lines from ChatGPT text
Sora Image Watermark Remover
Remove C2PA metadata watermarks from Sora generated images
Sora Image Watermark Detector
Detect C2PA metadata watermarks in Sora generated images
DALL-E Image Watermark Remover
Remove C2PA metadata watermarks from DALL-E generated images
DALL-E Image Watermark Detector
Detect C2PA metadata watermarks in DALL-E generated images