Beginner question: What does 'credits' actually mean? How many do I need?

I've been conducting a thorough evaluation of HeyGen for potential integration into our media asset pipeline, and while the technical documentation is generally adequate, I've found the core pricing and resource unit—the "credit"—to be somewhat opaque from an operational planning perspective. For those of us designing systems with predictable costs and load profiles, this ambiguity presents a significant hurdle.

Based on my analysis of their pricing tiers and my own experimentation, a "credit" in HeyGen's system is a non-fungible consumption unit applied differently depending on the media generation task. It is not a simple 1:1 mapping to seconds of output, which is a common initial assumption. The critical factor is the **output resolution and the inclusion of AI avatars**.

Here is a breakdown I've compiled from their documentation and API responses:

* **1 Credit** typically consumes 1 second of a generated video when using a **template-only** workflow (static imagery with voiceover). This is the baseline.
* **2 Credits** are consumed per second when generating video with an **AI Avatar** (a "Speaking Photo" or "Photo-to-video" asset). This effectively doubles the consumption rate.
* **4 Credits** are consumed per second when generating video with an **AI Avatar at 4K resolution**. This is the premium tier for high-fidelity output.

Therefore, the calculation for a project is not simply `duration in seconds = credits`. You must use the formula:
```
required_credits = video_duration_in_seconds * credit_multiplier
```
Where the `credit_multiplier` is:
- `1` for template videos
- `2` for AI Avatar videos (1080p)
- `4` for AI Avatar videos (4K)

**Practical Example:** To produce a 30-second explainer video using a custom avatar at 1080p, you would require `30 seconds * 2 credits/second = 60 credits`. The same video at 4K would require `120 credits`.

This leads to the second part of the question: "How many do I need?" From an architectural standpoint, this depends entirely on your expected throughput. If you are prototyping or producing short-form content intermittently, a starter pack may suffice. However, for any serious workflow aiming to integrate HeyGen into a microservice for, say, personalized video generation at scale, you must model your expected monthly output:
1. Estimate the total **duration** of video you plan to generate monthly.
2. Determine the **percentage** of that duration that will use avatars versus templates.
3. Apply the multipliers accordingly.

Failure to accurately model this will result in either unexpected credit exhaustion (causing workflow failures in an integrated system) or over-provisioning of resources. I strongly recommend setting up a monitoring layer in any integration to track credit consumption per job, mapping it to the `credit_multiplier` logic above, to maintain financial and operational predictability.