feat(video): add Alibaba Wan 2.6 text-to-video

[steebchen]

Summary

Adds video generation via Alibaba (DashScope) to the gateway, starting with the text-to-video model wan-2-6-t2v (wan2.6-t2v). This is the first video integration for the existing Alibaba provider; only text-to-video is wired up for now.

What changed

• packages/models — new wan-2-6-t2v model (output: ["video"], videoGenerations: true) with:
  • Per-second pricing: $0.08/s (720p) and $0.12/s (1080p)
  • Supported sizes 1280x720 / 720x1280 / 1920x1080 / 1080x1920 and durations 2–15s
  • Audio always on (supportsVideoAudio: true, supportsVideoWithoutAudio: false) — DashScope auto-dubs wan2.6 output and exposes no disable toggle
  • Mapping marked test: "skip" so it is excluded from the default e2e suite
• apps/gateway/src/videos/videos.ts — createAlibabaVideoJob submits an async DashScope task via POST /api/v1/services/aigc/video-generation/video-synthesis with the X-DashScope-Async: enable header and { model, input.prompt, parameters.{size,duration} } body (size converted to DashScope W*H format). Added the dispatcher case and the default base URL.
• apps/worker/src/services/video-jobs.ts — fetchAlibabaStatus polls GET /api/v1/tasks/{task_id}, maps output.task_status (PENDING/RUNNING/SUCCEEDED/FAILED) to the canonical status and surfaces the direct output.video_url. Alibaba is excluded from the /v1/videos/{id}/content metadata fetch since DashScope returns a direct OSS URL.

Defaults to the international endpoint https://dashscope-intl.aliyuncs.com (matching the existing Alibaba chat default). Uses the existing LLM_ALIBABA_API_KEY credential — no new env/provider setup required.

Testing

• pnpm format and pnpm build pass
• pnpm vitest run packages/models/src/providers.spec.ts apps/gateway/src/lib/alibaba-pricing.spec.ts — 153 passed
• E2E intentionally skipped (real paid video API)

Notes / follow-ups (out of scope)

• Image-to-video and region selection (the provider already declares Singapore/US/Beijing endpoints) are not wired up yet — text-to-video only, as requested.

🤖 Generated with Claude Code

Summary by CodeRabbit

• New Features

  • Added Alibaba (DashScope) as a video-generation provider.
  • New "Wan 2.6 Text-to-Video" model: up to 15s, multiple resolutions, audio support, per-second pricing.

• Content & Moderation

  • Recognition of Alibaba "green-net" moderation signals (improved content-filter matching).

• Tests

  • Added a test covering the new moderation signal handling.

• Integration

  • Worker/status polling and provider status handling updated to include Alibaba.

[coderabbitai]

Caution

Review failed

The pull request is closed.

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro

Run ID: f6daf99b-d81f-49f9-9d95-a776158be6ac

📥 Commits

Reviewing files that changed from the base of the PR and between a0d4809 and 1e9ef19.

📒 Files selected for processing (2)

• packages/shared/src/content-filter.spec.ts
• packages/shared/src/content-filter.ts

---

Walkthrough

Adds Alibaba (DashScope) video-generation support: registers a new video model, implements gateway-side async job creation and routing to Alibaba, and adds worker-side status polling that normalizes task status, progress, result URLs, and errors. Also updates content-filter signals and tests for Alibaba moderation messages.

Changes

Alibaba Video Generation Provider

Layer / File(s)	Summary
Model definition for Alibaba video generation packages/models/src/models/alibaba.ts	Defines wan-2-6-t2v model with video output type, 15s max duration, per-resolution per-second pricing, supported sizes and durations (2–15s), and audio support flags.
Gateway job creation and routing apps/gateway/src/videos/videos.ts	Adds alibaba base URL mapping, implements createAlibabaVideoJob to build DashScope async payloads (prompt, parameters), POST to video-synthesis endpoint, parse output.task_id, throw 502 on missing task id, and routes createUpstreamVideoJob to this handler.
Worker status polling and metadata handling apps/worker/src/services/video-jobs.ts	Adds alibaba base URL, isAlibabaVideoProvider, and fetchAlibabaStatus to query DashScope task status, normalize task_status/output into job metadata (status, progress, video URLs, mime, structured errors), integrates into fetchUpstreamStatus, and excludes Alibaba from generic content-metadata fetching.
Content filter signals and tests packages/shared/src/content-filter.ts, packages/shared/src/content-filter.spec.ts	Adds "Green net check failed" to content-filter signals and a Vitest case asserting detection of Alibaba DashScope “green-net” moderation text.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~20 minutes

Possibly related PRs

• theopenco/llmgateway#2486: Also adds entries to packages/models/src/models/alibaba.ts (related model-definition edits).
• theopenco/llmgateway#2270: Previously extended video-provider routing and status-polling infrastructure—related provider plumbing changes.
• theopenco/llmgateway#2114: Touched the Alibaba model registry; related at the model-definition level.

Suggested reviewers

• smakosh

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The PR title 'feat(video): add Alibaba Wan 2.6 text-to-video' accurately summarizes the main change—adding support for Alibaba's Wan 2.6 model for text-to-video generation across gateway and worker services.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch alibaba-wan-t2v-video

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: a0d4809657

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Honor Alibaba's configured region endpoint

When an Alibaba key is configured for us-virginia or cn-beijing via the existing LLM_ALIBABA_REGION/alibaba_region support, this new video path still submits to the Singapore endpoint. Alibaba documents that the endpoint URL and API key must belong to the same region (cross-region calls fail), so video generation will be broken for users whose chat/image Alibaba setup works in a non-default region. The worker has the same hard-coded default for polling, so the fix needs to resolve the provider's region endpoint consistently for creation and status checks.

Useful? React with 👍 / 👎.