Qwen-Image Is Here on SiliconFlow: Superior Text Rendering & Precise Image Editing

Sep 22, 2025

Today, Qwen-Image — a 20B MMDiT foundation model — is officially available on SiliconFlow. Built for both image generation and editing, it delivers major breakthroughs in complex text rendering, while also providing precise image editing performance and reliable all-around image capabilities.

Whether you're a content creator working with multilingual materials or a developer building text-integrated visual applications, Qwen-Image delivers the precision and flexibility to bring any creative vision to life.

With SiliconFlow's Qwen-Image API, you can expect:

Budget-Friendly Pricing: Qwen-Image $0.042/image.
Exceptional Text Rendering: Supports both alphabetic languages (e.g., English) and logographic languages (e.g., Chinese) with high fidelity.

Key Capabilities & Benchmark Performance

Unlike traditional T2I models that struggle with text rendering and easily break image consistency during detailed edits, Qwen-Image is designed to support:

Superior Text Rendering: Qwen-Image excels at complex text rendering, including multi-line layouts, paragraph-level semantics, and fine-grained details. It can generate complete documents, poster designs and other complex text layouts, ensuring accurate rendering from headlines to footnotes.
Consistent Image Editing: It achieves exceptional performance in preserving both semantic meaning and visual realism during editing operations. In practice, this means when users tweak images — for example, changing a product background, adding text to a poster, or refining details in a design draft — Qwen-Image keeps the rest of the image intact and natural, ensuring the edit blends seamlessly with the original picture.
Multi-style image generation: Qwen-Image supports a wide range of artistic styles—from photorealistic and impressionistic to anime and minimalist—making it a flexible tool for artists, designers, and storytellers.

Qwen-Image demonstrates strong multimodal capabilities, outperforming top models like GPT Image 1 and FLUX.1 Dev across image generation, editing and text rendering benchmarks:

On image generation, it scores 0.91 on GenEval and 88.32 DPG, leading the field.
For image editing, it tops GSO (15.11) and GEdit (EN: 7.56 /CN: 7.52).
In text rendering, especially in Chinese, it achieves state-of-the-art scores: 0.946 on LongText-ZH, 0.963 on OneIG-Render-ZH and 0.583 ChineseWord-ZH.

Real-world Performance on SiliconFlow

Text Rendering

Whether it's the wooden signboard of a traditional Japanese ramen shop or a modern bookstore exhibition in English, Qwen-Image renders the text accurately across languages.

A traditional Japanese street scene at evening, with a cozy ramen shop featuring a prominent wooden signboard with clear, accurate Japanese text "麺屋さくら".

Prompt: Generate a refined bookstore interior scene with warm lighting, wooden bookshelves, and an inviting reading atmosphere. On a central bookshelf, display three books Ocean Dreams, Midnight City and The Secret Garden as promotional highlights.

Multi-style image generation

The examples demonstrate image generations in Baroque, Makoto Shinkai style and traditional Chinese ink landscape painting, reflecting the model's deep understanding of various artistic styles and meticulous attention to detail.

Prompt: A beautiful landscape scene in Baroque style — dramatic chiaroscuro lighting, golden glow, ornate details, classical European architecture in the background, rich oil painting textures, highly detailed and majestic atmosphere.

Prompt: A beautiful landscape scene in Makoto Shinkai style — the cinematic anime look, vibrant colors, glowing sunset sky, shimmering reflections on water, soft clouds, emotional and dreamy mood, with subtle elements of modern life such as a distant train, city skyline, or power lines.

Prompt: A beautiful landscape scene in traditional Chinese ink painting style — monochrome ink wash, misty mountains, flowing river, pine trees, ancient pagoda, elegant brush strokes, large areas of blank space, minimalistic and poetic atmosphere.

Get Started Immediately

1. Explore: Try Qwen-Image in the SiliconFlow playground.
2. Integrate: Use our OpenAI-compatible API. Explore the full API specifications in the SiliconFlow API documentation.

import requestsurl = "https://api.siliconflow.com/v1/images/generations"payload = {    "batch_size": 1,    "num_inference_steps": 20,    "guidance_scale": 7.5,    "model": "Qwen/Qwen-Image",    "prompt": "an island near sea, with seagulls, moon shining over the sea, light house, boats int he background, fish flying over the sea"}headers = {    "Authorization": "Bearer <token>",    "Content-Type": "application/json"}response = requests.post(url, json=payload, headers=headers)print(response.json())

From text to stunning images, Qwen-Image brings ideas to life — and at SiliconFlow, we can't wait to see the amazing creations our community will make!

Ready to explore?

Business or Sales Inquiries →

Join our Discord community now →

Explore all available models on SiliconFlow →