DeepSeek-OCR compression in readable Rust (crates.io)

optical-embeddings v0.3.0 was published with a new component called DeepSeek-OCR — described as a way to “compress text into images” and implemented in readable Rust — but the package README failed to load on the registry and the release shows zero downloads so far. That makes the announcement effectively a lightweight code release without accompanying documentation or adoption data; developers who want to evaluate it will need to fetch the crate/source to inspect algorithms and examples directly. The idea is noteworthy for AI/ML because encoding text into images can enable novel retrieval, storage, and multimodal workflows: images carrying compressed textual payloads can be indexed by vision-based embedding pipelines, transported through image-centric delivery systems, or obfuscated/archived where binary/text channels are constrained. Implementing this in Rust suggests a focus on performance, memory safety, and easy integration into production services. Key technical questions for practitioners are the compression fidelity (lossy vs. lossless), OCR/decoding robustness across models and image degradations, compression ratio versus embedding density, and how image-based encodings interact with vision-language encoders. Given the missing README, the community should look for test vectors, decoding benchmarks, and security/privacy analyses before adopting DeepSeek-OCR in pipelines.