Discussions

Hi HeyGen Engineering Team,

We’re building a production pipeline that generates 3 Video Agent variants (A/B/C) per request. We need help confirming the exact request payload for Video Agent so we can:

1. Attach product image correctly:

• We understand Video Agent does not accept direct URLs in files[] and requires an asset_id from the Upload Asset API first.
• Please confirm the exact files[] schema for Video Agent (e.g., { asset_id, type, name/role, etc. }).

2. Lock avatar identity consistently:

• When no asset is included, Video Agent sometimes returns different/wrong avatars across A/B/C.
• We need to lock to a specific persona every time.
• Please confirm which fields Video Agent supports for identity locking (e.g., character_id, avatar_id, look_id) and provide a full working JSON example.

Context:

• We can retrieve avatars via GET /v2/avatars.
• We call Video Agent generate and pass our own callback_url (signed query param) for completion tracking.

If you provide a minimal working request JSON that guarantees (a) the correct avatar and (b) the correct product image attachment using asset_id, we’ll implement it immediately.

Thank you,
Rashid