Extract Stable Diffusion Metadata
and Local Workflows
Quick Summary (TL;DR)
DEUTLI is a 100% client-side parser for batch reading hidden tEXt chunks from generated PNG files (Stable Diffusion, ComfyUI, A1111). The tool locally extracts generation parameters, prompts, and node connection architecture, saving them into a structured .deut text format via the File System Access API without uploading images to a server.
The Fragility of tEXt Chunks
Generative AI networks store data about prompts, weights, seeds, and graphs directly in the PNG file header as binary text blocks (tEXt chunks). This architecture is critically vulnerable. When sharing a file through corporate messengers, social networks, or simply compressing it, these chunks are permanently deleted (metadata stripping).
For professionals in the visual arts, losing the pipeline architecture means the inability to accurately reproduce a result. DEUTLI solves this problem by allowing you to extract and isolate metadata before it is lost.
Client-Side Parsing vs. Cloud Extractors
Using third-party servers to read generation parameters compromises the intellectual property of design studios. Our architecture is built on complete Data Sovereignty:
- Zero-Server Architecture: The computational process occurs entirely within your browser. Not a single byte of commercial secrets is transmitted over the network.
- Batch Directory Processing: Thanks to the File System Access API, the parser works with folders, not single images. It scans hundreds of generations simultaneously, saving hours of routine work.
- Isolated Assets (Sidecar files): Chaotic JSON arrays and parameters are extracted and converted into standardized .deut files. They are saved directly to your hard drive next to the original PNGs, creating a reliable, software-independent backup.
- Frictionless Integration: The tool requires no Python environment setup, command line, or custom nodes. Don't type. Snap it in.
Frequently Asked Questions
How can I be sure you are not uploading my images to a server in the background?
Try the Proof of Trust test right now – disconnect the internet and check for yourself. Load the DEUTLI tool page in your browser, then completely disconnect your computer from the internet (unplug the cable or turn off Wi-Fi). Open a directory with your generations into the extractor. Ensure your browser allows 'File System Access' permission. Because the tool is built on a 100% client-side (zero-server) architecture and uses your PC's hardware power, it will successfully read the tEXt chunks and save the results locally even while entirely offline. Your files physically cannot leave your device.
How to safely extract a hidden prompt from an AI-generated image?
Point the client-side extractor to the directory containing your source PNG files. The parser will locally read the binary metadata within the browser and instantly create text files with generation parameters on your drive without sending images to a server.
Why is my ComfyUI workflow missing?
Workflows are stored as JSON strings inside the image. Saving images through third-party viewers or publishing them online often automatically strips these text blocks to optimize file size, leading to a complete loss of the node structure.
Why is the .deut format better than storing parameters in txt or json?
The .deut format is not just a local text file; it is a documented Open Standard. Unlike chaotic txt notes or unreadable machine json blobs, .deut standardization guarantees the long-term architectural compatibility of your metadata with the entire evolving ecosystem of prompt engineering tools. You are not locked into a specific generator, interface, or vendor – you get clean, structured data ready for versioning and seamless integration into any future visual arts pipelines.
Read .DEUT Open Standard on GitHub