Qwen Image 3.0 Pro Spicy API

Qwen Image 3.0 Pro Spicy is Alibaba's flagship third-generation image generation model, released with the Qwen-Image-3.0 series by the Qwen team on July 21, 2026, and built on a Diffusion Transformer (DiT) architecture, engineered for high information density and professional production. It accepts ultra-long prompts of up to 4,500 tokens, generating complex layouts like newspaper front pages, multi-panel storyboards, academic papers, and detailed UI interfaces in a single pass. It delivers micro-level detail — micro-expressions, pores, hair strands — with near-photographic fidelity, and industry-leading typography rendering legible small text down to 10px across 12 native languages and 20+ fonts. As of August 5, 2026, it ranked fifth globally on Arena.ai's Text-to-Image human-preference arena at 1263±11, the highest-ranked Chinese image generation model (Preliminary).

The model is offered through the iCreat platform as an API using a two-step asynchronous task flow: submit a task to obtain a task_id, then poll the result endpoint until a terminal status. It supports ten output sizes across the 1K and 2K tiers with differentiated per-image pricing: $0.04 per image at 1K and $0.075 at 2K; the output watermark can be turned off via parameter.

Model Positioning

Qwen Image 3.0 Pro targets flagship-tier professional image production. Unlike the standard Qwen Image 3.0, which balances speed, quality, and price, the Pro tier is purpose-built for complex content and fine rendering: long-text posters, dense infographics, multilingual layouts, UI interfaces, knowledge-heavy content, and first drafts of formal commercial work. Alibaba summarizes this generation as "rich content, authentic details, deep knowledge" — Pro is the full expression of that capability set, while Standard is the balanced variant for everyday batch production on the same architecture. On Arena.ai's human-preference Text-to-Image leaderboard, Pro ranks fifth globally and first among Chinese models at 1263±11, making it the go-to choice for complex layouts and small-text rendering.

Core Capabilities

Ultra-Long Complex Instructions

Prompts run up to 4,500 tokens — describe copy, multi-level headings, layout structure, module positions, chart data, and font/color requirements the way you would write a design brief. Newspaper front pages, multi-panel storyboards, and academic pages generate in one pass.

Micro-Level Detail Fidelity

Micro-expressions, pores, and hair strands are faithfully rendered with material and lighting fidelity approaching real photography — suited to formal commercial output of fine products and people.

Industry-Grade Multilingual Typography

Text renders legibly down to 10px across 12 native languages and 20+ fonts; small text stays legible, so text in multilingual posters, packaging, and infographics is usable as generated.

Dense Layouts and Interface Simulation

Supports dense nested layouts; mainstream web, game, and livestream interfaces can be simulated; product feature grids, knowledge cards, and selling-point graphics generate in a single pass.

Unified Generation and Editing

Text-to-image and image-to-image instruction editing share one model: add a reference image to content and edit by text instruction — outfit swaps, background changes, restoration, style transfer — with the flagship rendering capability carrying over to editing.

Pricing

Billed per generated image with differentiated 1K/2K pricing. The costUSD field in the response returns the actual cost once the task succeeds.

Application Scenarios

• Complex commercial posters: first drafts of formal layouts dense with long text, modules, and brand information
• Multilingual localized marketing: native text rendering for cross-language posters, packaging, and cross-border assets
• Publishing and education layout: production-grade newspaper pages, exam sheets, academic pages, and knowledge infographics
• UI and product prototyping: simulated generation of web, game, and livestream interfaces
• Instruction-based editing: outfit swaps, background changes, restoration, and style transfer

Model Comparison

Same-Series Comparison

Cross-Model Comparison

Same-series data comes from Alibaba's official releases; cross-model prices are the vendors' official API prices; this model's pricing and specifications follow the iCreat channel documentation.

Why Choose Qwen Image 3.0 Pro?

• 4,500-token instruction window plus dense nested layouts — complex pages generate in one pass, no stitching
• 10px small text across 12 languages and 20+ fonts: text in multilingual commercial assets is usable as generated, skipping post-layout
• Micro-expression, pore, and hair-strand level detail raises the fidelity floor for formal commercial output
• Fifth globally and first among Chinese models on Arena.ai's human-preference arena (1263±11, Preliminary) — proven on complex-layout tasks
• Unified generation and instruction editing, with flagship rendering applied to editing scenarios

Specifications

Architecture

Qwen Image 3.0 Pro is built on a Diffusion Transformer (DiT) architecture as the third-generation flagship of the Qwen image series. It carries the unified generation-editing design: text-to-image and image-to-image instruction editing share the same weights, so typography, micro-level detail, and world knowledge accumulated on the generation side carry directly over to the editing side. The instruction window extends to 4,500 tokens, specifically strengthened for high-information-density layouts, supporting nested layouts and composed multi-type visuals in a single generation. Output covers ten sizes across the 1K and 2K tiers with differentiated pricing.

Notes

• Submit and query must be chained with the same task_id
• While processing, result is []; keep polling, with a suggested interval of 2–5 seconds
• FAILED is terminal; check request parameters and reference media URLs before resubmitting
• input and parameters are sibling top-level fields; do not nest them
• Each item in messages[].content must be either {"text": "..."} or {"image": "..."}, combined as needed
• Reference image URLs must be publicly accessible
• parameters.size is required, written as width*height (e.g. 1024*1024), not joined with x
• watermark defaults to off; pass true explicitly when a watermark is needed
• costUSD is returned only when the task succeeds; failed tasks do not produce a cost field

FAQ

How do I get started with the Qwen Image 3.0 Pro API?​

Register on the iCreat platform and obtain an API Key from the console, send a generation request to the submit endpoint, then poll the result endpoint with the returned task_id. All requests are authenticated with the Authorization: Bearer header. Keep your API Key safe and never expose it in client code or public repositories.

How is the cost calculated?​

Billed per generated image with differentiated pricing: $0.04 per image at 1K and $0.075 per image at 2K. For example, generating 10 1K images costs $0.40, while 10 2K images cost $0.75. The costUSD field in the response gives the actual cost once the task succeeds.

Does it support generation or editing?​

Both. With only {"text": "..."} in content, it generates from the prompt; adding an {"image": "..."} reference enables instruction-based editing (outfit swaps, background changes, restoration, style transfer, and more). Generation and editing share the same model and endpoint.

What output sizes are supported?​

parameters.size supports the 1K tier (1024*1024, 1280*720, 720*1280, 1024*768, 768*1024) and the 2K tier (2048*2048, 2560*1440, 1440*2560, 2048*1536, 1536*2048) — ten values in total, with differentiated pricing between tiers. Pick the right size for each delivery platform.

How do I query the task result?​

Send a POST request to https://api.icreat.ai/v1/task/result with the task_id in the body. A status of SUBMITTED means processing, with result as [] — keep polling. When status is SUCCEEDED, result returns the image resource array; read url or download_url to retrieve the image.

What if the task fails?​

FAILED is a terminal status. Check in order: whether messages contains at least one text item, whether reference image URLs are publicly accessible, whether size is one of the ten values written as width*height, and whether input and parameters are sibling top-level fields. Fix any issue and resubmit. The cost field is returned only when the task succeeds.