SLATEMOTH / INSIGHT
Metering multimodal AI: text, image, video and audio
A modality-aware approach to measuring AI usage without forcing every workload into a token-only model.
The problem
Tokens are useful for many text models, but they are not a universal unit for AI services. Images, video, speech, embeddings and generated media can use different billable dimensions.
Text
Text usage can include input tokens, output tokens, cached tokens and provider-specific reasoning dimensions. Preserve original usage facts before commercial pricing is applied.
Image and video
Image workloads may depend on image count, dimensions, quality or operation type. Video may depend on duration, resolution, frame rate, audio inclusion or generation mode.
Audio
Speech-to-text is commonly measured by audio duration, while text-to-speech may be measured by characters, tokens or generated duration.
Measurement before pricing
Metering records what happened. Pricing decides what that usage costs. Keeping those stages separate lets provider cost and customer price evolve without rewriting historical usage.