Doubao Seedream Image Generation

Problem to solve

When image generation is added to an agent workflow, the friction is often not the model call itself but the surrounding plumbing: Ark API authentication, model version selection, parameter constraints, and saving the result to disk. Doubao Seedream also has a minimum pixel requirement, and using 2k, 4k, or WIDTHxHEIGHT is more predictable than writing ratios like 9:16. Doubao Seedream Image Generation keeps this flow in one script: the user provides a prompt, the skill applies sensible defaults, calls the API, downloads the image, and returns a local path.

How it works

• Parameter extraction: reads prompt, size, n, optimize, watermark, seed, output_path, and image; missing values fall back to defaults.
• Generation: runs scripts/generate_image.py with doubao-seedream-4-5-251128 by default, supporting text-to-image, image-to-image, and multiple outputs.
• Delivery: the script prints the saved path, which the agent can deliver to the user or surface through a preview surface.
• Prompt shaping: organizes descriptions as subject + action/state + environment/background + style/aesthetic, with standard prompt optimization available.

For sizing, the default 2k is about 2048x2048; use 2k for avatars, icons, and product shots, 4k for high-resolution assets, 1440x2560 for vertical posters, and 2560x1440 for banners or covers.

Boundaries and notes

The skill requires ARK_API_KEY and depends on volcengine-python-sdk[ark]. Seedream requires at least about 3.686 million pixels, so 1k and 1024x1792 do not qualify. Supported size formats include WIDTHxHEIGHT, 2k, and 4k; ratio notation such as 1:1 and 9:16 is not supported. Prompts should stay under 300 Chinese characters or 600 English words.