Am I the only one who prefers the text-to-speech over the avatars?

Having now conducted a comparative analysis of Synthesia's two primary output modalities—the generative avatars and the text-to-speech (TTS) audio engine—I find myself consistently opting to disable the video feed entirely and consume only the synthesized speech. This preference appears to be anomalous within my professional network, where the avatar technology is often cited as the core value proposition. I am curious if this forum contains individuals with a similar analytical bent who have arrived at the same conclusion.

My rationale is rooted in bandwidth efficiency, information density, and cognitive load. The avatars, while technologically impressive, introduce a significant amount of visual data that, in my assessment, carries minimal informational value relative to its cost (in attention, bandwidth, and processing). In a typical technical explainer video, the semantic content is almost entirely contained within the narration, with the avatar's lip movements and limited gestures serving as redundant reinforcement at best, or a distracting uncanny valley effect at worst.

Consider a practical deployment scenario: generating internal training modules for a distributed engineering team. The functional requirements are the accurate conveyance of procedural knowledge and architectural concepts. When A/B testing internally, we observed no statistically significant improvement in knowledge retention when comparing avatar-present versus TTS-only versions of the same script. However, we measured notable differences in:
* **Production Latency:** TTS-only workflows completed approximately 40-60% faster, as they bypassed the computationally intensive and queue-prone video render pipeline.
* **Iteration Speed:** Correcting a script error or updating a fact required a full video re-render with the avatar path, versus a sub-5-minute TSS re-generation.
* **Bandwidth & Storage:** The video files were orders of magnitude larger, complicating distribution and archiving.

Furthermore, from a pure audio fidelity perspective, Synthesia's TTS engines (particularly the premium voices) demonstrate exceptional prosody, technical pronunciation accuracy, and lack of the artifacting that can sometimes manifest in the vocal track when it is forced to synchronize perfectly with generated lip movements. There is a measurable, if subtle, degradation in audio quality when the two streams are locked together.

I posit that for use cases where authority and factual accuracy are paramount over charismatic presentation—such as internal communications, detailed API documentation, or data-dense briefings—the TTS output alone is not just sufficient, but optimal. It strips the process down to its most efficient form: script in, high-fidelity audio out. The avatar layer, in these contexts, functions as a costly visual wrapper that appeals more to managerial stakeholders than to the end-user's ability to absorb information.

Is this a perspective born from my specific focus on technical content, or have others found that beyond the initial novelty, the avatar's utility diminishes rapidly for substantive material? I am particularly interested in any structured evaluations or benchmarks others may have conducted comparing engagement metrics between the two modalities for non-marketing content.