Qwen Image 2.0: Fast Text-to-Image AI with 2K Resolution

Qwen Image 2.0 — Text to Image

Qwen Image 2.0 is Alibaba's unified image generation model, running a 7B-parameter architecture that delivers native 2K resolution and built-in typography rendering. This standard endpoint is optimized for speed, making it ideal for rapid prototyping, prompt exploration, and iterative workflows at $0.035 per image.

Standard vs Pro

The standard tier prioritizes generation speed for fast iteration. When you're ready for final production assets with maximum detail and text accuracy, switch to the Pro endpoint ($0.075/image).

Key Parameters

Parameter	Default	Range	Notes
prompt	—	up to 1,000 tokens	Describe subject, style, and composition
negative_prompt	—	string	Exclude unwanted elements
image_size	square	enum or custom	`square`, `square_hd`, `landscape_4_3`, `landscape_16_9`, `portrait_4_3`, `portrait_16_9`
guidance_scale	5	0–20	Lower = creative; higher = strict adherence
num_inference_steps	28	1–50	15–20 for drafts, 25–30 for quality
seed	random	integer	Fix for reproducible outputs
num_images	1	1–4	Batch multiple images in one request
output_format	png	png / jpeg / webp	Choose based on file size needs

Quick Start

python
import fal_client

result = fal_client.subscribe(
    "fal-ai/qwen-image-2/text-to-image",
    arguments={
        "prompt": "Watercolor illustration of a cozy Japanese ramen shop at night, warm lantern glow, rain-slick street reflections, Studio Ghibli atmosphere",
        "image_size": "landscape_4_3",
        "num_inference_steps": 25,
        "guidance_scale": 5,
    }
)

print(result["images"][0]["url"])

javascript
import { fal } from "@fal-ai/client";

const result = await fal.subscribe("fal-ai/qwen-image-2/text-to-image", {
  input: {
    prompt: "Watercolor illustration of a cozy Japanese ramen shop at night, warm lantern glow, rain-slick street reflections, Studio Ghibli atmosphere",
    image_size: "landscape_4_3",
    num_inference_steps: 25,
    guidance_scale: 5,
  },
});

console.log(result.data.images[0].url);

Iteration Workflow

A practical approach for finding the right output:

Explore — generate at 15–20 steps to test prompt ideas quickly
Refine — bump to 25–28 steps once you have a direction
Finalize — switch to the Pro endpoint at 35+ steps for production renders

Use a fixed `seed` while adjusting prompt wording to isolate what each change does.

Prompting Tips

Front-load subjects: "Red sports car on mountain road, aerial drone shot, golden hour" beats "A nice landscape photo with a car somewhere."
Be specific about style: "editorial photography", "watercolor illustration", "low-poly 3D render" — avoid vague terms like "beautiful" or "amazing."
One style per prompt: "Photorealistic oil painting" confuses the model. Pick one direction.
Negative prompts help: `"blurry, distorted, deformed, watermark, text artifacts"` cleans up common issues.

Best Use Cases

Concept exploration — quickly test ideas before committing to Pro-quality renders
Social media content — fast turnaround for posts and stories
Mood boards and references — generate visual directions for creative projects
Batch generation — produce multiple variations cheaply to find the best composition

Related Endpoints

Text to Image (Pro) — higher fidelity for final assets
Image Editing — modify existing images with natural language
API Reference — full parameter documentation

fal-ai/qwen-image-2/text-to-image

Input

Result

What would you like to do next?

Logs