ERNIE-Image

SiliconFlow ERNIE-Image API Complete Guide: The Fast Text-to-Image Service at $0.015/Image

Jun 28, 2026

SiliconFlow ERNIE-Image API Complete Guide: The Fast Text-to-Image Service at $0.015/Image

Summary: SiliconFlow has officially launched Baidu's ERNIE-Image-Turbo model API at $0.015/image (¥0.11). This is the fourth third-party API platform for ERNIE-Image, following FAL.AI, Atlas Cloud, and WaveSpeed AI. This article covers everything from basic API calls to batch generation techniques and platform comparisons, helping you choose the best deployment option.

SiliconFlow Launches ERNIE-Image-Turbo

SiliconFlow has officially listed Baidu's ERNIE-Image-Turbo as its 163rd model on the platform. Notably, this deployment was completed independently by Baidu's ERNIE-Image team — leveraging SiliconFlow's elastic GPU service to significantly reduce the time from resource preparation to service launch.

According to SiliconFlow, the elastic GPU service provides an "out-of-the-box" production-level model deployment experience, greatly simplifying the process of deploying models into highly available inference service endpoints. This means ERNIE-Image-Turbo's powerful image generation capabilities can be quickly shared with the platform's tens of millions of users.

Key details:

  • Model: ERNIE-Image-Turbo (8B parameters, 8-step inference)
  • API price: ¥0.11/image (~$0.015/image)
  • New user credit: ¥16 / $1
  • Model ID: baidu/ERNIE-Image-Turbo

Getting Started with API

SiliconFlow uses a standard OpenAI-compatible API format, making ERNIE-Image-Turbo calls straightforward.

Python Example

import http.client
import json

api_key = "your-siliconflow-api-key"


payload = {

"model": "baidu/ERNIE-Image-Turbo",

"prompt": "A cinematic photograph of a sunset on a city rooftop, a young woman wearing a khaki trench coat holding a camera",

"size": "1024x1024",

"n": 1

}


headers = {

"Authorization": f"Bearer {api_key}",

"Content-Type": "application/json"

}


conn = http.client.HTTPSConnection("api.siliconflow.cn")

conn.request("POST", "/v1/images/generations",

body=json.dumps(payload), headers=headers)

response = conn.getresponse()

result = json.loads(response.read().decode())

print(result)

cURL Example

curl -X POST "https://api.siliconflow.cn/v1/images/generations" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "baidu/ERNIE-Image-Turbo",
    "prompt": "A cinematic photograph of a sunset on a city rooftop",
    "size": "1024x1024",
    "n": 1
  }'

Supported Parameters

Parameter Type Description
model string Model ID: baidu/ERNIE-Image-Turbo
prompt string Positive prompt (supports Chinese and English)
size string Output size, recommended: 1024x1024, 1024x768, 768x1024
n integer Number of images to generate, default 1
response_format string Return format: url (default) or b64_json (Base64)

Typical Use Cases

ERNIE-Image-Turbo on SiliconFlow is particularly well-suited for:

Poster Design: Complex text layouts and multi-element compositions — Turbo mode completes high-quality generation in 8 steps.

Comic Storyboarding: Character consistency and multi-panel layouts, where ERNIE-Image's structured layout capabilities excel.

Product Image Batch Production: E-commerce product photos and social media content — low latency directly translates to productivity gains.

Interactive Creation: Real-time creative workflows where 8-step inference latency makes the "prompt → visual result" cycle extremely smooth.

Comparison with Other API Platforms

Platform Price Model Version Features
SiliconFlow ¥0.11/image Turbo Deployed by Baidu team, new user credit
FAL.AI ~$0.015/image Base/Turbo + LoRA Supports cloud LoRA training
Atlas Cloud Pay-per-use Base/Turbo Official API, batch production optimized
WaveSpeed AI Commercial pricing Base/Turbo Enterprise deployment, batch generation API

SiliconFlow's advantages include transparent pricing, a low barrier for new users with free credits, and service stability backed by the original model team.

Batch Generation Best Practices

For batch generation scenarios, SiliconFlow recommends:

  1. Control concurrency: Avoid too many simultaneous requests — 1-2 concurrent requests recommended
  2. Use URL response format: Default URL return is ideal for async processing
  3. Base64 format: Set response_format: "b64_json" for direct app embedding
  4. Size selection: Extreme aspect ratios reduce quality — stay close to 1:1 or 4:3

SiliconFlow's Model Ecosystem

SiliconFlow currently hosts 160+ models covering language, image, audio, and video tasks. Beyond ERNIE-Image-Turbo, it also supports Z-Image-Turbo, Qwen-Image, and other mainstream text-to-image models. Developers can flexibly call different models through a single API key, achieving "Token Freedom."

Sources: SiliconFlow official documentation, Zhihu column, Baidu Qianfan Community, SiliconFlow API documentation

ERNIE-Image Team