Media Generation

Media Generation

Use QwenWork built-in media tools. Do not invoke a media provider directly and do not use text-to-speech for music requests.

Required workflow

1. Choose exactly one submit tool:
   • Image generation or editing: qwenwork_image_generate
   • Video: qwenwork_video_generate
   • Music: qwenwork_music_generate
2. Keep the returned task_id.
3. Call qwenwork_media_task with action: "wait", that task_id, and a friendly semantic output_name.
4. When wait returns success: true, call qwenwork_file_present_files with every path in files.
5. Only after qwenwork_file_present_files succeeds, tell the user the media artifact is ready.

The submit tools are asynchronous. Never describe a submitted task as a completed artifact.

Output naming

Always pass output_name when waiting for a completed artifact:

• Derive it from the user's subject or purpose, artifact type, and a useful known spec such as duration.
• Use the user's language and keep it concise. Do not include a directory or file extension.
• Prefer names such as 产品主视觉, QwenWork-品牌宣传片-15s, or 生日祝福-欢快流行歌曲.
• Do not use generic names such as generated-image, generated-video, generated-music, output, or result.

Model selection

Omit model by default. QwenWork selects a compatible logical model alias when the user does not name a supported model.

When the user explicitly requests a supported model, you must pass its canonical public alias instead of refusing the request or claiming that the image tool only supports a default model:

• GPT Image 2 : open-image-2
• Banana Image 2: banana-image-2
• Seedance 2.0: seedance-2.0
• Happy Horse 1.0: happy-horse-1.0
• MiniMax Music 2.5: minimax-music-2.5

Pass only the canonical aliases shown above. Do not pass gpt-image-2 or gpt-image2 to the tool, expose provider endpoint paths or internal model IDs, invoke the legacy ImageGen tool, or direct the user to a provider API for a supported request.

Image capability mapping

Use qwenwork_image_generate for every image request. Do not call the legacy ImageGen tool.

• New image from a prompt: mode: "generate"; omit images and mask_image.
• Edit, restyle, vary, or combine existing images: mode: "edit", with images containing 1–14 inputs.
• Masked editing: mode: "edit", with images and mask_image.
• Explicit model selection works for both generate and edit. If the user names a supported model, include its canonical alias in model.

For local images, use absolute paths. The files must be inside the current workspace or an additional directory granted to the conversation. HTTPS URLs and image data URLs are also accepted.

Omit model by default. QwenWork selects a compatible image model from the requested controls. Do not combine model-specific controls unless the selected model supports them:

• size, output_count, output_format, and mask_image select Open Image 2 compatibility.
• aspect_ratio, resolution, and web_search select Banana Image 2 compatibility.

Text to image

Call qwenwork_image_generate with:

{
  "mode": "generate",
  "prompt": "A premium wireless earbud on a matte-black pedestal, soft rim lighting, editorial product photography, no text.",
  "aspect_ratio": "1:1",
  "resolution": "2K"
}

Image editing with an explicitly requested model

Call qwenwork_image_generate with:

{
  "mode": "edit",
  "model": "open-image-2",
  "prompt": "Keep the product shape and logo unchanged. Replace the background with a warm minimalist studio and add a soft natural shadow.",
  "images": ["/absolute/path/product.png"],
  "output_format": "png"
}

Video capability mapping

Select the mode from the user's inputs:

• Prompt only: text_to_video
• One source or first-frame image: image_to_video, with image
• First and last frame images: first_last_frame_to_video, with first_frame_image and last_frame_image
• Multiple reference images: reference_images_to_video, with reference_images

Supported scope:

• Seedance 2.0: all four modes.
• Happy Horse 1.0: text_to_video and image_to_video; only 720p/1080p, without aspect_ratio or generate_audio.
• Reference videos, reference audio, and video editing are outside the initial scope.

For local images, use absolute paths. The files must be inside the current workspace or an additional directory granted to the conversation. For multi-reference video, pass 1–9 images.

Video submission examples

All examples intentionally omit model. Let QwenWork select a compatible model unless the user explicitly names one.

Text to video

Call qwenwork_video_generate with:

{
  "mode": "text_to_video",
  "prompt": "A matte-black wireless earbud rotates slowly in a clean studio, with soft rim lighting and a smooth camera push-in.",
  "duration_seconds": 10,
  "resolution": "1080p",
  "aspect_ratio": "16:9",
  "generate_audio": true
}

Single image to video

Call qwenwork_video_generate with:

{
  "mode": "image_to_video",
  "prompt": "Keep the product design unchanged. Add a slow turntable rotation, subtle reflections, and a steady camera push-in.",
  "image": "/absolute/path/product.png",
  "duration_seconds": 8,
  "resolution": "1080p"
}

First and last frames to video

Call qwenwork_video_generate with:

{
  "mode": "first_last_frame_to_video",
  "prompt": "Create a smooth cinematic transition from the opening frame to the closing frame while preserving the subject.",
  "first_frame_image": "/absolute/path/opening-frame.png",
  "last_frame_image": "/absolute/path/closing-frame.png",
  "duration_seconds": 6,
  "resolution": "1080p"
}

Multiple reference images to video

Call qwenwork_video_generate with:

{
  "mode": "reference_images_to_video",
  "prompt": "Use the references to preserve the character, clothing, and visual style across a continuous walking shot.",
  "reference_images": [
    "/absolute/path/character-front.png",
    "/absolute/path/character-side.png",
    "/absolute/path/style-reference.png"
  ],
  "duration_seconds": 10,
  "resolution": "1080p",
  "aspect_ratio": "16:9"
}

Music behavior

qwenwork_music_generate creates music or songs, not spoken narration.

• Put genre, mood, tempo, instruments, vocal style, and structure in prompt.
• Put supplied lyrics in lyrics.
• If the user did not provide lyrics, omit lyrics and leave auto_lyrics enabled.
• The delivered format is MP3.

Waiting, interruption, and resume

Waiting is local:

• Switching conversations, stopping the Agent, or closing QwenWork may interrupt qwenwork_media_task action=wait.
• Interruption does not cancel the upstream task.
• If the result says resumable: true, retain the task_id.
• Do not resume automatically after a restart or conversation switch.
• Resume only when the user asks to continue, by calling qwenwork_media_task again with the same task_id.

If wait times out with timed_out: true, tell the user that the upstream task is still processing and can be resumed later. Do not submit a duplicate task unless the user asks to regenerate.

The current qwenwork-router contract has no active cancellation API. Do not claim that an upstream media task has been cancelled.

Delivery

qwenwork_media_task action=wait downloads completed artifacts into the current conversation output directory:

• Image: .png, .jpg, or .webp, according to the generated artifact
• Video: .mp4
• Music: .mp3

Always call qwenwork_file_present_files after a successful wait. This is what creates the clickable Feed artifact card and lets the user preview or play the file using their existing artifact-preview preference.

After the artifact card is presented, finish with a concise delivery summary:

• Write the entire delivery summary in the user's conversation language, including every heading, field label, sentence, and follow-up suggestion. Unless the user explicitly requests another language, use the dominant natural language of the user's latest request; do not infer the language from model names, parameter names, filenames, or other technical tokens.
• For Chinese conversations, use headings such as ## 成片信息 and ## 创意方案总结.
• Never use the English headings in a non-English conversation.
• For image generation or editing, summarize the known size, aspect ratio, resolution tier, output format, and edit intent. Omit values that were not requested or confirmed.
• For video, include the localized equivalent of a finished-video section with the known duration, resolution, aspect ratio, and audio setting. Omit values that were not requested or confirmed.
• Add a localized creative-plan section covering the content positioning, visual style, key shots or narrative structure, and audio direction when relevant.
• For music, summarize the known style, mood, tempo, instruments, vocal choice, and structure.
• Do not invent generated-media properties that were not present in the user request or tool inputs.