[RelUnrelated]

Hi everyone,

I've just updated the main post with the Version 1.1.0 release of the AI Vision Metadata plugin. This is a major under-the-hood architectural update that significantly expands the plugin's capabilities and overall stability, especially for those of you running large batches.

Here are the major highlights for this release:

Native OpenRouter Integration: You can now route requests through OpenRouter's massive library of vision-capable models. The plugin dynamically fetches the model list directly from their API—you don't even need an API key entered just to browse what's available! I've also set the default fallback to openrouter/auto for built-in load balancing.

Dynamic Anthropic Fetching: The hardcoded model lists are gone. The plugin now utilizes Anthropic's new /v1/models endpoint, meaning any future Claude vision models will appear in your dropdown automatically the moment they are released by Anthropic.

Server Resilience (Exponential Backoff): If you process large library batches, you shouldn't see any more UI crashes from Google 503 traffic errors or 429 rate limits. The plugin now silently catches these network hiccups and uses an exponential backoff routine to wait and retry without interrupting your queue.

Local LLM Hardening: For those of you running local vision models via LM Studio or Ollama, I've implemented a custom JSON sanitizer. It automatically strips out the conversational padding or markdown that local models sometimes inject, preventing parsing crashes.

The new AI_Vision_Metadata_v1.1.0.zip is attached to the first post in this thread. Be sure to check the updated README inside the archive for configuration details, especially regarding the new provider options.

As always, let me know if you run into any edge cases or unusual behavior with the new routing logic!