iptag/jimeng-api

Overview

This skill generates images using a locally deployed Jimeng API service (Dreamina). It supports text-to-image and image-to-image workflows with selectable models, aspect ratios, and resolutions, and automatically saves outputs to the project /pic folder. Always request the user's Session ID before running any generation.

How this skill works

The skill calls the Jimeng HTTP API (default http://localhost:5100) via a wrapper script that runs text or image generation and downloads returned image URLs. It only passes parameters the user explicitly requests, relies on built-in defaults otherwise, converts WebP to PNG, timestamps filenames, and reports saved file paths. The script supports multiple models, intelligent ratio detection for compatible models, and up to 10 input images for image-to-image.

When to use it

• User asks to generate artwork, illustrations, or visual content using Jimeng/Dreamina.
• User requests text-to-image or image-to-image transformation with specific prompts.
• User needs custom ratio or resolution (1k/2k/4k) and model selection.
• You have a local Jimeng API (Docker) running and can provide a Session ID.
• User asks to save generated images to the project /pic folder.

Best practices

• Always ask for and verify the Session ID before proceeding; require region prefixes for international sites (us-, hk-, jp-, sg-).
• Only pass parameters the user explicitly requests; otherwise rely on defaults (ratio 1:1, resolution 2k, model jimeng-4.0).
• Clarify whether the task is text-to-image or image-to-image and request input images or URLs when needed.
• Use intelligent ratio only with jimeng-4.0, 4.1, or 4.5 models; other models ignore it.
• Inform the user of exact output file paths after saving; do not open or analyze generated images in-session.

Example use cases

• Generate a 4K 21:9 landscape scene for a wallpaper using jimeng-4.0 and --resolution 4k --ratio 21:9 (only if user requests those flags).
• Transform a provided photo into oil-painting style via image-to-image: user supplies image path or URL and a style prompt.
• Create multiple variations from a single prompt; script downloads all returned images and lists saved PNG paths.
• Auto-select vertical/horizontal ratio using --intelligent-ratio for prompts that mention 'portrait' or 'landscape' (with compatible models).
• Run nanobanana model on an international site (us-/hk-/jp-/sg-) with known resolution constraints.