Best practices for video pre-processing (resolution) with media_resolution parameter in Gemini 3.0 (

Gemini API
,
1

Hi everyone,

I have a question regarding the optimal video resolution when using the new Gemini 3.0 models, specifically concerning the media_resolution parameter.

Context: Previously, with Gemini 2.5 Pro/Flash, the community consensus for “Best Practice” was to pre-process videos to 1 FPS and low resolution (e.g., 360p) to maximize file efficiency without losing model performance, as the model consumed a fixed number of tokens per frame regardless of the input file’s quality.

My Question: With the introduction of the media_resolution parameter in Gemini 3.0, I am curious about the internal behavior when media_resolution is set to medium (or low).

  1. Internal Downscaling: If I upload a 1080p video but set the API call to media_resolution="medium", does the API internally transcode/downscale this video to a specific resolution (e.g., 480p) before processing?

  2. Efficiency: If internal downscaling occurs, would uploading a 1080p file be considered a waste of bandwidth?

  3. Best Practice: For the medium setting, is it still considered a best practice to pre-resize the video to 360p or 480p on the client side before uploading?

I want to ensure I am not uploading unnecessarily large files if the model is only going to “see” a low-resolution version anyway.

Any insights on the internal target resolutions for each media_resolution tier would be greatly appreciated.

Thanks!

3

Hi @user3473 , Welcome to the AI Forum!!!

Thanks for reaching out to us.

The media_resolution parameter determines the maximum number of tokens allocated per input image or video frame, allowing you to balance response quality against latency and cost.

  1. Internal Downscaling: For video, media_resolution_medium is capped at 70 tokens per frame to optimize context usage and video is compressed more aggressively. This 70 tokens per frame requires downsampling to fit the 1080p resolution data into the limited visual context for the model.

  2. Efficiency: Since both media_resolution_low and media_resolution_medium are capped at 70 tokens per frame, uploading a 1080p file might be inefficient and considered as a waste of bandwidth.

  3. Best Practice: To optimize efficiency, you can consider pre-resizing the video to 360p or 480p before uploading.

Please refer to this official documentation on Media Resolution here: Media resolution