huajing-image-gen

Generate images through a local Huajing (AI Image Studio) server. Use when the user asks to create or generate an image, polish an image prompt, or look up previously generated images. Supports text-to-image with polling, batch count, size, negative prompt, and provider/model selection. Requires the local Huajing server (default http://127.0.0.1:4173) and Node 18+.

Huajing Image Generation

Drive a locally running Huajing (AI Image Studio) server to generate images. Huajing manages your own image-model providers (OpenAI-compatible relays, chat-based image adapters, etc.) and stores every generated image in a local gallery, so generation is cheap to retry and fully auditable.

Setup

1. Start the Huajing server (single instance per machine):
   • Windows: run start.bat in the Huajing repo, or npm start.
   • Any OS: npm start (listens on 127.0.0.1:4173 by default).
2. Add at least one image provider in the Huajing UI (Service & Models page): base URL + API key + image model.
3. Optional environment overrides:
   • HUAJING_BASE_URL — server base URL, default http://127.0.0.1:4173.
   • HUAJING_DATA_DIR — Huajing data/ directory; when set, results include absolute local file paths instead of only HTTP URLs.

Usage

All commands print JSON on stdout and human-readable diagnostics on stderr. Exit codes: 0 success, 1 generation failed, 2 server unreachable, 3 bad usage.

Check the server before doing anything else:

bash

node {baseDir}/scripts/huajing.mjs health

Generate an image (blocks until the image is done, polling internally):

bash

node {baseDir}/scripts/huajing.mjs generate \
  --prompt "a tiny red paper crane on a wooden desk, soft window light" \
  --size 1024x1024 --count 1 --timeout 300

Options: --prompt (required), --size (WxH, default 1024x1024), --count (1–4), --provider (provider id or case-insensitive name; default: first enabled provider with an image model), --model (override the provider's default image model), --negative (negative prompt), --timeout seconds (default 300), --plain (human-readable output instead of JSON).

Polish a rough prompt into a richer one using the provider's prompt model:

bash

node {baseDir}/scripts/huajing.mjs polish --prompt "一只纸鹤" --provider <id-or-name>

Search the local gallery for previously generated images:

bash

node {baseDir}/scripts/huajing.mjs gallery --limit 5 --query "crane"

Notes

• If generate exits with code 2, the server is not running — tell the user to start it with start.bat / npm start in the Huajing repo.
• If generation fails with a provider error (exit code 1), the error text from the upstream model service is included in the JSON error field; retrying with a different --provider usually helps.
• All generated images are kept in the local gallery (gallery subcommand), so you can re-reference earlier results without regenerating.