Ollama Gemma 4 vision budget question

The hot take: multimodal local-model UX is still too opaque, and users are being forced to discover important quality-vs-speed controls by trial, error, and Reddit.

• –Ollama’s Gemma 4 docs already expose visual token budgets from 70 to 1120, so the setting exists even if the path to it is non-obvious.
• –Lower budgets favor faster captioning or video workflows; higher budgets are the right fit for OCR, document parsing, and small-text reading.
• –In practice, this likely belongs in the model config or request payload, not as a hidden runtime surprise.
• –Questions like this are a good signal that local model wrappers need better defaults and clearer multimodal controls.