API

Image API

Paid accounts: POST /v1/images/generations (JSON) and POST /v1/images/edits (multipart image + prompt). Temporary price: 1 credit per output image; n: 1–3, steps: 1–60. PNG URLs require authentication and expire after 15 minutes. Use response_format=b64_json for embedded output, or DELETE /v1/images/files/{image_id} after downloading a URL. POST /v1/images/quote returns the price before generation; available sizes are in /v1/models → image_formats.

curl https://api.ghostmetal.io/v1/images/generations -H "Authorization: Bearer $GHOSTMETAL_KEY" -H "Content-Type: application/json" -d '{"model":"qwen-image-2.1","prompt":"A watercolor fox","size":"1024x1024","n":1,"negative_prompt":"blur"}'

negative_prompt works only when true_cfg_scale > 1. If a nonempty negative_prompt is provided without true_cfg_scale, it defaults to 2. Explicit true_cfg_scale=1 is preserved. Both parameters remain available in the API.

Chat completions

POST /v1/chat/completions — streaming (SSE) and non-streaming, OpenAI-compatible.

from openai import OpenAI

client = OpenAI(
    base_url="https://api.ghostmetal.io/v1",
    api_key="YOUR_GHOSTMETAL_KEY",
)

stream = client.chat.completions.create(
    model="glm-5.3",
    messages=[{"role": "user", "content": "Write a streaming API client in Python"}],
    stream=True,
)
for chunk in stream:
    print(chunk.choices[0].delta.content or "", end="")

Speech API

POST /v1/audio/speech Generate audio for a paid account. Requests are free and counted in usage; the model must be enabled and its backend configured. Trial accounts can use browser speech.

curl https://api.ghostmetal.io/v1/audio/speech \
  -H "Authorization: Bearer $GHOSTMETAL_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"piper","input":"Привет, мир!","voice":"alloy","response_format":"opus","normalization":"ru"}' \
  --output speech.ogg

input: up to 5,000 characters; voice: accepted and ignored (one female voice); response_format: opus (Ogg/Opus), mp3 (default), or wav; speed: 0.5–2. Each request is read in one voice: English words are spoken in Russian with at least as many Cyrillic as Latin letters. Russian text is always normalized (numbers, units, English terms); the optional normalization field (none, ru, ru_translit) is accepted for compatibility.

The response contains audio bytes. One request per account runs at a time in each gateway process; a new waiting request replaces the previous one. This endpoint supports a subset of the OpenAI API.

Models & pricing

Billed in credits from your balance. See tokenomics. Moderation is OFF on API keys (except the Business/Safe SKU). Rate limit: 120 req/min per key.