In our previous post, we detailed the mechanics of metadata storage in generated PNGs and the concept of the open .deut format. The fragility of node dependencies and the resulting data loss are well-documented, so let's cut right to the chase.
We have completely rewritten the architecture and released DEUTLI Extractor V2. What started as a basic client-side parser has evolved into an independent, industrial-grade utility for managing metadata in isolated environments.
What's Under the Hood in V2
Zero-Install (Portable Binaries): We've eliminated the need for local developers to wrestle with Python environment conflicts. V2 ships as compiled, independent executables for three major OS families: Windows (.exe), macOS (.dmg), and Linux (.AppImage). The utility requires no installation or administrator privileges—it runs straight from a removable USB drive.
Visualization and Batch Mode: You no longer need to drag and drop files one by one. The built-in Thumbnail Viewer instantly scans your selected directory (even if it contains thousands of heavy PNGs), rendering a grid of your generations. You can extract data selectively or trigger an asynchronous batch mode for the entire folder.
100% Zero-Telemetry: The tool is built for closed studio perimeters (air-gapped workflows). The core runs exclusively on the user's local hardware (CPU/GPU). There are no trackers, network modules, or background pings. Your source files never leave your local drive.
Deep Parsing and Sidecar Philosophy: The compute core recursively traverses ComfyUI node graphs and Automatic1111/Forge text matrices. The original binary remains untouched. Instead, the extractor drops a lightweight, structured.deutsidecar file next to it with an identical filename, ready for integration into corporate databases.
Frequently Asked Questions (FAQ) for V2
Q: Do the portable versions of V2 require an internet connection?
A: Absolutely not. The tool is fully autonomous. You can physically unplug your ethernet cable—the batch extraction process will run locally, guaranteeing 100% NDA compliance.
Q: How fast is the batch mode on large directories?
A: The architecture is optimized for heavy files. Internal benchmarks show that on standard hardware, parsing a folder of 1,000 images (2-3K resolution) and generating individual .deut sidecar files takes about 7 minutes.
Q: What happens to the old web version of the extractor?
A: It has been updated to V2 and operates as a PWA (Progressive Web App) at extractor.deut.li. Once the page loads, the interface caches and functions entirely offline. However, for massive end-to-end parsing of entire directories, we highly recommend using the portable desktop binaries.
Q: Can I embed your parser into my own script?
A: Yes. We are strictly open-source. The entire codebase is distributed under the MIT License. You can pull our repository, conduct an independent security audit, compile the application yourself, or extract the parsing logic to integrate into your own custom pipelines.
Links:
Download Portable Builds: GitHub Releases
Source Code: GitHub Repository
Format Specification: deut.li