Zenith-AI API

Reference documentation for inference endpoints

POST /v1/chat/completions/stable

Chat completion endpoint. Accepts an OpenAI-compatible message array and returns the full completion object.

Headers

HeaderTypeDefaultDescription
Authorization* string— Must be present and follow the format Bearer <token>. Requests without a valid authorization header are rejected with 401.

Request Body

FieldTypeDefaultDescription
messages* array— Array of message objects (role + content). Supports multimodal content arrays.
modelopt string— Model name. Defaults to "Auto".
temperatureopt float0.0 Sampling temperature (0–2).
top_popt float0.95 Nucleus sampling probability.
max_tokensopt integernull Maximum tokens to generate.
streamopt booleanfalse Enable streaming responses.

Example Request

curl -X POST https://inference.zenith-ai.one/v1/chat/completions/stable \
  -H "Authorization: Bearer <your-token>" \
  -H "Content-Type: application/json" \
  -d '{
    "messages": [
      {"role": "system", "content": "You are a helpful assistant."},
      {"role": "user",   "content": "Explain black holes in one sentence."}
    ],
    "temperature": 0.7
  }'

Response

Standard OpenAI chat completion object.

{
  "id": "chatcmpl-...",
  "object": "chat.completion",
  "choices": [{
    "index": 0,
    "message": {"role": "assistant", "content": "..."},
    "finish_reason": "stop"
  }],
  "usage": {"prompt_tokens": 20, "completion_tokens": 18, "total_tokens": 38}
}
POST /v1/image/generation/stable

Image generation endpoint. Supports text-to-image and image-to-image (editing/variation) via an optional images field. Returns the raw GenerateContentResponse with inline image bytes base64-encoded.

Headers

HeaderTypeDefaultDescription
Authorization* string— Must be present and follow the format Bearer <token>. Requests without a valid authorization header are rejected with 401.

Request Body

FieldTypeDefaultDescription
prompt* string— Text description of the image to generate.
modelopt stringgemini-2.5-flash-image Must be one of: gemini-2.5-flash-image, gemini-3-pro-image-preview.
response_modalitiesopt array["IMAGE","TEXT"] Modalities to return. Use ["IMAGE"] to suppress text.
temperatureopt float— Sampling temperature.
top_popt float— Nucleus sampling probability.
top_kopt float— Top-k sampling cutoff.
seedopt integer— Fixed seed for reproducible outputs.
max_output_tokensopt integer— Maximum tokens in the response.
safety_settingsopt array— List of SafetySetting objects to override default filters.
system_instructionopt string— System-level instruction prepended to the conversation.
candidate_countopt integer— Number of response candidates to generate.
stop_sequencesopt array— Strings that stop generation when encountered.
imagesopt array[] Input images for image-to-image generation. Each item is either:
• An object {"data": "<base64>", "mime_type": "image/png"}
• A data-URI string "data:image/png;base64,<base64>"
Supported MIME types: image/png, image/jpeg, image/gif, image/webp. Invalid base64 returns 400.

Example — Text to Image

curl -X POST https://inference.zenith-ai.one/v1/image/generation/stable \
  -H "Authorization: Bearer <your-token>" \
  -H "Content-Type: application/json" \
  -d '{"prompt": "a futuristic city at sunset, photorealistic"}'

Example — Image to Image (editing)

curl -X POST https://inference.zenith-ai.one/v1/image/generation/stable \
  -H "Authorization: Bearer <your-token>" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "Change the sky to a dramatic purple sunset",
    "images": [
      {"data": "<base64-encoded PNG>", "mime_type": "image/png"}
    ]
  }'

Or using a data-URI:

{
  "prompt": "Make the background a beach",
  "images": ["data:image/jpeg;base64,<base64-encoded JPEG>"]
}

Response

Raw Google GenerateContentResponse. Image bytes are base64-encoded inside candidates[].content.parts[].inline_data.data. 

{
  "candidates": [{
    "content": {
      "role": "model",
      "parts": [
        {
          "inline_data": {
            "data": "<base64-encoded image bytes>",
            "mime_type": "image/png"
          }
        },
        {
          "text": "Here is the generated image."
        }
      ]
    },
    "finish_reason": "STOP"
  }],
  "usage_metadata": { "prompt_token_count": 9, "candidates_token_count": 0 }
}

Decoding the Image (Python)

import requests, base64

# Text-to-image
resp = requests.post(
    "https://inference.zenith-ai.one/v1/image/generation/stable",
    headers={"Authorization": "Bearer <your-token>"},
    json={"prompt": "a cat sitting on a cloud"},
).json()

# Image-to-image (pass base64-encoded source image)
with open("source.png", "rb") as f:
    b64_image = base64.b64encode(f.read()).decode()

resp = requests.post(
    "https://inference.zenith-ai.one/v1/image/generation/stable",
    headers={"Authorization": "Bearer <your-token>"},
    json={
        "prompt": "make the background a snowy mountain",
        "images": [{"data": b64_image, "mime_type": "image/png"}],
    },
).json()

for i, candidate in enumerate(resp.get("candidates", [])):
    for j, part in enumerate(candidate["content"]["parts"]):
        if part.get("inline_data"):
            img_bytes = base64.b64decode(part["inline_data"]["data"])
            open(f"image_{i}_{j}.png", "wb").write(img_bytes)
        elif part.get("text"):
            print("Text:", part["text"])