Choosing a model

Updated Oct 11, 2026 · 1 min read

Pick a model from the provider you chose, and optionally include thumbnails in prompts if the model can see images.

Generation model, in Settings → AI tagging, lists the models of the provider you picked. OpenRouter's list is its live catalog, with a solid default already selected. OpenAI, Anthropic and Google Gemini start on a preset, and you can type any other model ID from that provider, click Use model ID and test it. Ollama lists the models installed on this computer, and the Codex CLI offers Codex default, which lets Codex choose, plus the models your Codex account lists.

Out of the box, tagging works from text: the model reads a file's name, folder path, and existing tags, and writes tags plus a searchable description from those.

There's also an optional “Include thumbnails in prompts” switch. Turn it on and Meshory attaches a small copy of each model's rendered thumbnail, which sharpens tags considerably for cryptically-named files. It requires a model that accepts images: the switch only activates when the model you've picked does. A model ID Meshory doesn't know stays text-only, and every Codex model listed takes thumbnails.

Beyond that it's a trade-off you control: stronger models produce better tags and descriptions but cost more per model; lighter models are cheaper for tagging a huge library. Settings shows an estimated cost for your whole library with the model you've picked. A provider you pay with a key bills you per use, Ollama costs nothing but your computer's time, and the Codex CLI counts runs against your ChatGPT plan, so you decide where to land.