Hi there, I’m having the following issues with Gemini using the Vertex API.
Anyone came across the same issue or knows how to resolve?
Fundamental Gemini Video Processing Limitation
After extensive testing with all three Gemini models, we’ve discovered a critical and consistent limitation in Gemini’s video
processing:
Results Summary:
| Model | Processing Time | Duration Analyzed | Content Analysis |
|-----------------------|-----------------|---------------------|-------------------|
| gemini-2.5-flash-lite | ~6 min | ~6 minutes | OBS tutorial only |
| gemini-2.5-flash | ~48 min | ~16 minutes (1000s) | OBS tutorial only |
| gemini-2.5-pro | ~26 min | 2.3 minutes (138s) | OBS tutorial only |
Critical Findings:
1. ALL Gemini models truncate video analysis regardless of:
- Processing timeout (300s vs 60s)
- Response size limits (5MB vs 500KB)
- Schema validation limits (50 vs 10 thumbnails)
- Model capabilities (Pro vs Flash vs Lite)
2. Gemini 2.5 Pro performed WORST - only analyzing 138 seconds vs Flash’s 1000 seconds
3. The complete 36-minute ai.mp4 file is correctly generated (2190 seconds, 237MB)
4. All models only see initial OBS setup content, never the main music production lecture
Root Cause:
This appears to be an undocumented limitation in Google’s Gemini API where video processing stops at various arbitrary durations
despite:
-
Models claiming 40+ minute video support in documentation
-
Extended processing times suggesting full analysis
-
Correct file preparation and optimal configuration
This is a fundamental API limitation that cannot be fixed through configuration changes or different models.
The issue may be on Google’s side - either a bug in their video processing pipeline or an undocumented quota/limit that truncates analysis
regardless of processing time or model capacity.