media-craft（媒体工艺）

Media Craft — obra CLI Skill

Overview

obra is a unified CLI for AI media generation. It handles images, videos, and music through a consistent async-task workflow. Use this skill whenever the user asks to generate, create, edit, transform, upscale, or enhance any media.

Prerequisites

npm install -g obra
obra config set kie.apiKey YOUR_API_KEY

Core Workflow

Every obra command follows the same pattern:

1. Preferred: synchronous with --wait --json — blocks until the task completes and returns JSON with output URLs.
2. Save output directly — add -o <path> to download the result to a local file.
3. Async fallback — if --wait is not used, obra returns a task ID. Poll with obra status <id> --wait --json, then download with obra download <id> -o <dir>.

Always use --wait --json unless the user explicitly wants async behavior.

Image Generation

Generate images from text prompts.

obra image generate "a serene mountain lake at sunset" --wait --json

Key Flags

Model-specific parameters vary by model. Common ones include aspect_ratio, seed, negative_prompt, width, height. Use obra image info <model> to see available parameters for a specific model.

Example: specific model and aspect ratio

obra image generate "cyberpunk cityscape" --model google/imagen4 --param aspect_ratio=16:9 --wait --json

Available Image Models

Discover models with:

obra image list
obra image info <model-name>

Notable models: flux-2/pro-text-to-image, google/imagen4, grok-imagine/text-to-image, gpt-image/1.5-text-to-image, seedream/4.5-text-to-image, qwen/text-to-image, ideogram/character, z-image.

Image Editing & Transformation

Image-to-Image (Edit with Reference)

Provide an input image URL plus a prompt describing the edit. Use --param image_urls=<url> to pass the reference image (must be a URL, not a local path):

obra image generate "make it look like winter" --model seedream/4.5-edit --param image_urls=https://example.com/input.jpg --wait --json

Edit-capable models: seedream/4.5-edit, bytedance/seedream-v4-edit, qwen/image-edit, grok-imagine/image-to-image, gpt-image/1.5-image-to-image, google/nano-banana-edit, google/pro-image-to-image, ideogram/character-edit.

Use obra image info <model> to check the exact parameter names for each model.

Upscaling

Some upscale models take a task_id from a previous generation task:

obra image generate "" --model grok-imagine/upscale --param task_id=<previous-task-id> --wait --json

Other upscale models accept image URLs directly:

obra image generate "upscale" --model topaz/image-upscale --param image_urls=https://example.com/low-res.jpg --wait --json

Upscale models: grok-imagine/upscale, topaz/image-upscale, recraft/crisp-upscale.

Background Removal

obra image generate "" --model recraft/remove-background --param image_urls=https://example.com/photo.jpg --wait --json

Reframing (Extend / Outpaint)

obra image generate "extend the scene" --model ideogram/v3-reframe --param image_urls=https://example.com/photo.jpg --wait --json

Character Remix

obra image generate "character in a new style" --model ideogram/character-remix --param image_urls=https://example.com/character.jpg --wait --json

Video Generation

Generate videos from text or images.

obra video generate "a cat playing piano" --wait --json

Key Flags

Model-specific parameters (aspect ratio, duration, etc.) vary by model. Use obra video info <model> to see available parameters.

Image-to-Video

obra video generate "bring this scene to life" -i scene.jpg --wait --json

Available Video Models

obra video list
obra video info <model-name>

Notable models by provider:

• xAI: grok-imagine/text-to-video, grok-imagine/image-to-video
• Kuaishou: kling-2.6/text-to-video, kling-2.6/image-to-video, kling/v2-1-master-text-to-video
• OpenAI: sora-2-text-to-video, sora-2-pro-text-to-video, sora-2-image-to-video, sora-2-pro-image-to-video
• Alibaba: wan/2-6-text-to-video, wan/2-6-image-to-video
• MiniMax: hailuo/02-text-to-video-pro, hailuo/2-3-image-to-video-pro
• Bytedance: bytedance/seedance-1.5-pro, bytedance/v1-pro-image-to-video

Video Editing & Transformation

Video-to-Video

Transform an existing video with a new prompt:

obra video generate "convert to anime style" -i input.mp4 --model wan/2-6-video-to-video --wait --json

Video Upscaling

obra video generate "upscale" -i low-res.mp4 --model topaz/video-upscale --wait --json

Watermark Removal

obra video generate "remove watermark" -i watermarked.mp4 --model sora-watermark-remover --wait --json

Motion Control

obra video generate "camera pan left" -i scene.jpg --model kling-2.6/motion-control --wait --json

AI Avatars

Generate talking-head avatar videos:

obra video generate "introduce yourself as a tech reviewer" -i avatar.jpg --model kling/ai-avatar-pro --wait --json

Models: kling/ai-avatar-standard, kling/ai-avatar-pro.

Speech-to-Video

obra video generate "speaking character" -i audio.wav --model wan/2-2-a14b-speech-to-video-turbo --wait --json

Storyboard

obra video generate "scene 1: ...; scene 2: ..." --model sora-2-pro-storyboard --wait --json

Music Generation

Generate music from text prompts.

obra music generate "upbeat electronic dance track" --wait --json

Key Flags

Instrumental Track

obra music generate "calm lo-fi hip hop beat for studying" --instrumental --wait --json

Custom Mode

obra music generate "lyrics here..." --custom-mode --style "indie folk" --title "Morning Light" --wait --json

Lyrics Generation

Generate lyrics from a prompt (synchronous, no --wait needed):

obra music lyrics "a song about coding at 3am" --json

Music Videos

Generate a music video from a completed music task:

obra music video <taskId> <audioId> --wait --json

The taskId and audioId come from the output of a prior obra music generate command.

Synchronized Lyrics (Timestamps)

Get word-level timestamps for generated music:

obra music timestamps <taskId> <audioId> --json

Model Discovery

List and inspect available models for any media type:

obra image list          # list image models
obra video list          # list video models
obra music list          # list music models

obra image info <model>  # detailed info about a specific model
obra video info <model>

Task Management

obra status              # show all running tasks
obra status <id> --wait --json  # wait for a specific task
obra download <id> -o ./output  # download task results
obra history list        # list past tasks
obra history show <id>   # show details of a past task

Best Practices

1. Always use --wait --json to get structured output with download URLs in a single call.
2. Use -o <path> to save files directly when the user wants a local file.
3. Write detailed prompts — more descriptive prompts produce better results across all media types.
4. Chain workflows — e.g., generate an image, then use it as input for image-to-video, then upscale the video.
5. Check model availability — run obra <type> list to see current models before specifying one.
6. Parse JSON output — when --json is used, parse the JSON to extract URLs, task IDs, and metadata for downstream use.