tencent-hunyuanvideo

What it does

Separates requests into three contract families (TokenHub HY-Video-1.5, legacy vclm 3.0, and HunyuanVideo-1.5 local weights) and manages the end-to-end process for running Tencent HunyuanVideo tasks. It enforces a dry-run preflight, requires explicit approval with region, model, input, and output details, and uses a deterministic approval digest to claim and track submissions before execution. It supports planning 720p outputs via a provided example, with strict handling of prompts, optional image inputs, and result delivery constraints.

How it works

• Identifies three concrete surface options: TokenHub HY-Video-1.5, Legacy vclm, and HunyuanVideo-1.5 local.
• Before paid work or asset/model actions, it requires explicit disclosure of surface, model/checkpoint family, region, mode, input details, target resolution, audio plan, and QA gates.
• Enforces a plan-first workflow: defaults to plan-only, does not enable postpaid billing or automatic downloads.
• Uses a dry-run TokenHub client: constructs a payload with model "hy-video-1.5", prompt, and resolution 720p; can attach an image URL; computes a plan digest and outputs a plan with an approval_digest. Execution proceeds only when environment variables (TENCENT_APPROVED_DIGEST, TENCENT_TOKENHUB_API_KEY) and thresholds are satisfied.
• Submits to TokenHub endpoints via HTTP POST to /v1/api/video/submit and queries via /v1/api/video/query, handling response IDs and status updates until completion.
• Validates result URLs, enforces result host suffixes, and hashes the downloaded MP4 to verify integrity.

When to use it

Use when planning Tencent HunyuanVideo deployments that require strict preflight, region entitlement checks, explicit cost ceilings, and auditable submission flows for text-to-video or image-to-video tasks.

What it can touch

• TokenHub HY-Video-1.5 endpoints: https://tokenhub.tencentmaas.com and https://tokenhub-intl.tencentmaas.com
• Endpoints: "/v1/api/video/submit" and "/v1/api/video/query" (via HTTP POST)
• Payload fields include: model, prompt, resolution, image (optional URL), and logo_add
• Environment variables: TENCENT_APPROVED_DIGEST, TENCENT_APPROVE_MAX_CNY, TENCENT_TOKENHUB_API_KEY
• Local artifact handling for downloaded results (hashing, size checks, and storage paths)

Caveats

• Requires explicit entitlement and current price verification for the chosen region/site before execution.
• Uses a plan digest to prevent mismatched resubmissions; state is persisted to a .job.json alongside the output file to prevent duplicate submissions.
• The sample plan enforces 720p output and a 5-second result constraint for the TokenHub path; 1080p is not enabled unless the current schema and entitlement reflect it.
• AI disclosure and labeling are governed by default logo_add=1; changing this requires approval per Tencent guidance.
• Prices and quotas are subject to change; re-fetch price and billing state before submission.