Skip to content

Drop in a screenshot and convert it to clean code (HTML/Tailwind/React/Vue)

Python

Calibrate video token estimation from actual usage data

Abi Rajad580428da5a8
diff --git a/backend/video/cost_estimation.py b/backend/video/cost_estimation.py
index ddfab8d7..fdf014e2 100644
--- a/backend/video/cost_estimation.py
+++ b/backend/video/cost_estimation.py
@@ -26,22 +26,18 @@ class CostEstimate:
output_tokens: int
-# Gemini 3 video token rates per second (from documentation)
-# https://ai.google.dev/gemini-api/docs/media-resolution#token-counts
-# Note: Based on observed actual token counts, billing appears to use ~1 FPS
-# even when higher fps is requested. This may vary - adjust if actuals differ.
+# Gemini 3 video token rates per second (calibrated from actual usage)
+# Docs say 280 tokens/frame at HIGH + 32 audio = 312/sec, but actual usage
+# shows ~556 tokens/sec. This suggests ~2 FPS effective sampling or different rates.
VIDEO_TOKENS_PER_SECOND = {
- MediaResolution.LOW: 70,
- MediaResolution.MEDIUM: 70, # Same as LOW for video
- MediaResolution.HIGH: 280,
+ MediaResolution.LOW: 140, # ~2x the documented 70
+ MediaResolution.MEDIUM: 140,
+ MediaResolution.HIGH: 556, # Calibrated from actual: (7375-1846)/(11.17-1.22)
}
-# Audio tokens per second
-AUDIO_TOKENS_PER_SECOND = 32
-
-# System prompt + user text tokens (approximate)
-# Based on actual usage: ~1400-1500 tokens for typical prompts
-PROMPT_TOKENS_ESTIMATE = 1500
+# Prompt overhead (system prompt + user text)
+# Calibrated from actual usage: tokens - (duration * video_rate)
+PROMPT_TOKENS_ESTIMATE = 1200
# Pricing per million tokens (USD)
# https://ai.google.dev/gemini-api/docs/pricing
@@ -77,18 +73,13 @@ def get_model_api_name(model: Llm) -> str:
def estimate_video_input_tokens(
video_duration_seconds: float,
media_resolution: MediaResolution = MediaResolution.HIGH,
- include_audio: bool = True,
) -> int:
- # Estimate based on observed actual token counts (appears to bill at ~1 FPS)
+ # Calibrated from actual usage data (includes video frames + audio)
tokens_per_second = VIDEO_TOKENS_PER_SECOND[media_resolution]
- visual_tokens = int(video_duration_seconds * tokens_per_second)
-
- audio_tokens = 0
- if include_audio:
- audio_tokens = int(video_duration_seconds * AUDIO_TOKENS_PER_SECOND)
+ video_tokens = int(video_duration_seconds * tokens_per_second)
- # Add prompt tokens (system + user text)
- total_tokens = visual_tokens + audio_tokens + PROMPT_TOKENS_ESTIMATE
+ # Add prompt overhead (system prompt + user text)
+ total_tokens = video_tokens + PROMPT_TOKENS_ESTIMATE
return total_tokens
@@ -146,7 +137,6 @@ def estimate_video_generation_cost(
input_tokens = estimate_video_input_tokens(
video_duration_seconds=video_duration_seconds,
media_resolution=media_resolution,
- include_audio=True,
)
output_tokens = estimate_output_tokens(