
Integrating catalog · not Live media
AI video generation API: track Wan / MiniMax / HappyHorse without lying about Live
Why this topic is a status page, not a media SDK
Publishing fake Live video endpoints burns trust faster than ranking. The useful artifact is a shared catalog of Integrating IDs, family hubs, and engineering rules that refuse Chat Completions for frames.
When status flips to Live, this page should grow media docs and billing units — not pretend the chat route suddenly returns MP4s.
Catalog status
Family comparison
| Dimension | Wan | MiniMax Video-01 | HappyHorse |
|---|---|---|---|
| Primary job | Text-to-video demand around Qwen | Video-01 media + separate text agents | Named Integrating media row |
| Hub home | /models/qwen | /models/minimax | /models/qwen |
| Status today | Integrating | Integrating | Integrating |
| Do not | Call via Chat Completions for frames | Mix with M2/M3 agent budgets | Queue customer renders |
| Ship instead | Live DeepSeek text + feature flag | Live text + media flag | Demand tracking only |
Wan (Qwen hub)
Wan text-to-video demand is why many “Qwen API” searches are really video searches. On the Qwen hub, treat Wan as Integrating media inventory — separate from qwen-plus text evals.
Open hubMiniMax Video-01
MiniMax also sells an agent-coding story (M2/M3). Do not put Video-01 into the same retry and cost alerts as text agents. The MiniMax hub splits those intents on purpose.
Open hubHappyHorse
HappyHorse is catalogued beside the Qwen vendor filter as Integrating. Use it for demand tracking and naming consistency, not for a production render queue.
Open hubEngineering rules until Live
01
Do not point Chat Completions at video IDs
Expecting frames from a chat model string is how prototypes lie in demos. Wait for a documented media endpoint.
02
Keep the critical path on Live text
Product logic, agents and support copy stay on DeepSeek Live routes. Video is a feature flag behind status checks.
03
Budget media separately
When Live ships, token-style text budgets will not map cleanly to seconds or frames. Plan a separate wallet line item.
What to ship now
- Live DeepSeek text for agents, copy and control planes.
- UI copy that says video is coming / Integrating — never “available now.”
- Feature flags keyed off GET /v1/models + this page’s status table.
- Watchlists for Wan / MiniMax / HappyHorse demand without queueing jobs.
Internal and external links
Internal links keep decisions on SmarToken. External links are upstream references — reconcile both before you forecast spend or cut production traffic.
Internal links
- Qwen hub
Wan / HappyHorse inventory home
- MiniMax hub
Video-01 vs text agents
- DeepSeek hub
Live text while waiting
- SiliconFlow alternative
Platform decision twin
- 中文版
Chinese status mirror
External references
- Alibaba Model Studio video docs
Upstream Wan / video reference shape
- MiniMax docs
Vendor media + agent documentation
FAQ
- Can I render Wan or MiniMax video through SmarToken today?
- No as Live production media. The catalog lists Integrating rows so product and SEO teams share one status source. Wait for Live + media docs before customer jobs.
- Why publish a video page before the endpoint works?
- English search demand for AI video generation APIs clusters on Chinese families. A thin “coming soon” spam page fails; a status + decision page that refuses Chat Completions for frames is useful.
- Will video share /v1/chat/completions?
- Unlikely as the long-term shape. Text chat stays on Chat Completions. Media APIs usually need dedicated endpoints, async jobs and different billing units. Confirm docs when status flips.
- What should engineering ship while waiting?
- Keep agents, copy and control planes on Live DeepSeek text. Design the product so video is a feature flag, not a hard dependency in the critical path.