Doubao Image Generation

Problem

This skill targets a concrete workflow gap: you already have an OpenAI-compatible Doubao Seedream relay, but you do not want to hand-roll images/generations requests, convert reference images to base64, download remote files, retry network errors, or expose a clean result path for downstream scripts. It is intended for relay-based image generation, not direct official Ark API access.

How it works

The skill wraps image generation in a callable script. After configuring ARK_BASE_URL, ARK_API_KEY, model, and save directory, gen.py can be invoked from a CLI or function and prints a single JSON object that is easy to parse with json.loads. The main capabilities are:
- Text-to-image: pass a prompt and size; for the default Seedream model, use 2K or larger to avoid upstream minimum-pixel failures.
- Image-to-image: accept public URLs or local file paths; local images are converted to base64 before submission, and multiple references can be mixed.
- Sequential images: use n to generate a set of related images and return multiple local paths.
- Multi-image fusion: feed several reference images into the model to produce a fused image or a fused series.

Configuration and boundaries

Before calling the API, verify that api_key exists; if it is missing, ask the user to provide it or run the setup flow. By default, generated images are downloaded locally while the remote URL is also preserved; if download is disabled, only the URL is returned. Different relays may rename models or enforce different size constraints. Generated image URLs often expire after roughly 24 hours, so persistence should rely on local files. This skill is not intended for reading, analyzing, editing existing images, video generation, or speech tasks.