Models & downloads

What downloads, how big, and the Hugging Face token for speed.

On-device AI needs model weights. OY.Surf downloads them on demand and shows real progress (X / total GB). You can keep working while they download.

What downloads

• Description / search — a vision-language model + embedding model.
• Generation — Z-Image Turbo (text-to-image / image-to-image).
• Instruction editing — Qwen-Image-Edit (~27 GB), downloads on first click.
• Dictation — a Whisper model for Surf Flow.

Faster downloads

Add a Hugging Face token in Settings → AI models for faster, parallel downloads. A stalled partial shows as paused with a Resume button.

Where they live

Weights are cached locally and reused; once downloaded, the feature works offline.