feat: Native YouTube URL support for Gemini models (visual processing)

[pollinations-ai]

Reported by: mano.resells.

Context
Currently, when a YouTube URL is passed to Pollinations with a Gemini model, it appears to process it as text/web scraping (likely transcript only).

However, the native Google Gemini API supports processing YouTube videos directly (visual frames + audio) without manual frame extraction, often via file_uri or specific video input handling.

Feature Request

• Detect when a YouTube URL is passed to a Gemini model (specifically 1.5 Pro/Flash).
• Pass the URL to the Gemini API using the native video handling method to enable full multimodal analysis (visuals, actions, objects).
• This would allow prompts like "What color is the car?" or "Describe the action in the video" to work via Pollinations.

Ref
User noted that native Gemini API handles this efficiently without external downloading.

---

Author: _dr_misterio_

Created via Discord

[pollinations-ai]

Update from mano.resells.:

Please refer to the official Gemini API Video Understanding Documentation for implementation details.

Key point: The Gemini API supports passing YouTube URLs directly via file_uri in generateContent (currently in preview), which enables native visual analysis without manual frame extraction.

Example snippet:

types.Part(
    file_data=types.FileData(file_uri='https://www.youtube.com/watch?v=...')
)

---

From: mano.resells.

Added via Discord