Qwen

Qwen (/media/qwen) serves Alibaba's Qwen-Image and Wan2.x video models through the DashScope / Model Studio API. It handles long prompts and produces native 2K images plus text-to-video and image-to-video clips.

Where to configure

The Qwen media key is shared with the Qwen LLM provider. Configure your DashScope API key in Settings → AI (or Settings → Media) and both surfaces work. See Settings for provider setup.

Tabs / operations

Tab	Operation	Key fields	Output
Text → Image	Image	Model (qwen-image-plus/qwen-image), prompt, negative prompt, size, prompt-extend toggle, watermark toggle, optional seed	Still image
Text → Video	Video	Model (Wan 2.2/2.1 T2V variants), prompt, negative prompt, resolution, duration, prompt-extend toggle	MP4 video clip
Image → Video	Video	Model (Wan 2.2/2.1 I2V variants), source image, prompt, negative prompt, prompt-extend toggle	MP4 video clip

Generation flow

Completed artifacts land in /files via the Render Queue. Images finish synchronously (with brief internal polling); videos run asynchronously and are picked up by the media-jobs poll cron. From the queue card you can Edit in Designer or Post. See Media Studios overview for the shared flow.

Caveats

• Uses the DashScope async task API (X-DashScope-Async: enable on creation, then GET /tasks/{id}).
• The adapter routes prompt, negative_prompt, and img_url into DashScope's input object; every other parameter goes into parameters.
• There is no completion webhook; video jobs rely on the poll cron.
• 1080p resolutions require a Plus Wan model.
• Qwen-Image handles long prompts (up to roughly 1,000 tokens), so detailed scene descriptions are supported.

---

Verified against v1.0.0