Qwen Image 3.0: The Essentials
Last updated: August 6, 2026
Provider: Alibaba | Modality: Text to Image and Image to Image | Models: Qwen Image 3.0 and Qwen Image 3.0 Pro
Qwen Image 3.0 is Alibaba's image generation family on Scenario. Both tiers take a detailed prompt (and optionally 1 to 3 reference images) and return stills that hold up for film, ads, game key art, and dense text layouts such as posters, UI mockups, and multilingual headlines. Pro is the higher-fidelity tier with sharper detail up to 2K.
Which Model Should I Use?
Model | Best for | Notes |
|---|---|---|
Everyday concepts, exploration, strong text layouts | Same 1K and 2K size menu as Pro | |
Final-quality marketing, film stills, game key art | Higher fidelity; sharper detail at 2K sizes |
Pick Base when you are iterating. Switch to Pro when the frame needs to ship as a final still. The examples below were generated with Qwen Image 3.0 Pro.
How to Use the Model
Open either model, enter a detailed Prompt, set Size, and Generate. Leave Reference Images empty for text to image. Add 1 to 3 images when you want an edit, restyle, or composite.
Film, ads, and game stills (Qwen Image 3.0 Pro)
Write a full cinematic paragraph: subject, lighting, lens language, and what must stay out of frame. Turn Prompt Extend off when you already specified the shot.
A neo-noir film still: a lone detective in a long coat stands under a flickering streetlamp on rain-soaked cobblestones, deep shadows, wet reflections of neon red and teal, smoke drifting, 35mm anamorphic bokeh, cinematic color grade, tense quiet before a chase, no signage readable.
Neo-noir film still · Size 2688×1536 · Open on Scenario
A high-end perfume advertising hero shot: a faceted crystal bottle catching a single hard key light, golden liquid, floating glass shards and mist around it, black seamless backdrop, luxurious commercial still-life, specular sparkle, aspirational beauty campaign look, no labels with readable words.
Perfume campaign still · Size 1536×2688 · Open on Scenario
A dark fantasy game boss arena: a colossal armored knight with a flaming greatsword faces a player-scale rogue across cracked obsidian floor, cathedral ruins and ash rain, volumetric firelight, epic action-RPG key visual, painterly-photoreal hybrid, intense silhouette drama, no health bars or menus.
Dark fantasy boss arena · Size 2688×1536 · Open on Scenario
Text in images (Qwen Image 3.0 Pro)
Spell every headline, label, and language string in the Prompt. Quote the exact text you want to see. Works for posters, UI mockups, packaging, and multilingual calligraphy.
A vertical cultural poster with bold Chinese calligraphy as the hero. Large centered black brush characters: "山高水长". Smaller English translation under them: "Mountains high, rivers long". Soft ink-wash mountain silhouette behind the characters, pale rice-paper texture, red seal stamp lower-right reading "文心". Minimal contemporary gallery-poster layout, exquisite brush detail, high cultural visual impact.
Chinese calligraphy poster with English subtitle · Size 1536×2688 · Open on Scenario
A photoreal smartphone held vertically in one hand against a soft bokeh cafe background, screen fully visible. The phone shows a fitness app home screen titled "Pulse Run" in bold white at the top. Center: a circular ring progress graphic labeled "Weekly goal 78%". Below: three list rows with icons reading "Morning jog · 5.2 km", "Interval set · 24 min", "Recovery walk · 2.1 km". Bottom tab bar labels: "Home", "Plan", "Coach", "Me". Clean modern iOS-like UI, crisp sans-serif type, subtle glass card shadows, product photography lighting, high visual clarity for mobile UX showcase.
Mobile fitness UI with legible labels · Size 1536×2688 · Open on Scenario
Edits with reference images (Qwen Image 3.0 Pro)
Pass 1 to 3 Reference Images. Order matters: the model sees them in the order you add them. Name what to keep from each reference and what to change.
Combine both references: keep the detective character from the first image walking through the rainy neon cyberpunk street canyon from the second image. Preserve coat, hat, and silhouette; preserve wet asphalt reflections and dense futuristic buildings. Moody neon-noir crossover, anamorphic bokeh, no readable billboards or UI.Before Source detective still · Open on Scenario | After  Detective into cyberpunk canyon · 2 refs · Open on Scenario |
Parameters
Both models share the same controls. Parameter names below match the Scenario UI.
prompt
Required. Describe the image or the edit. Chinese and English are supported. Cap is 2500 characters. The film, text, and edit examples above all use a single detailed paragraph.
images
Optional. Reference Images, 1 to 3. Omit for text to image. Used in the detective-into-cyberpunk edit above.
negativePrompt
Optional. Content that should not appear. Useful for blocking accidental text, logos, or UI when you want a clean film still.
size
Output resolution as width×height. Presets cover 1K and 2K aspect ratios such as 1024×1024, 1344×768, 1536×2688, and 2688×1536. Film examples used 2688×1536; poster and UI examples used 1536×2688.
numOutputs
Number of images per run, 1 to 6. Default is 1.
promptExtend
When true, rewrites short prompts before generation. Set false when your prompt is already fully specified, as in the examples above.
promptExtendMode
direct or agent. Agent is text to image only and is unavailable when Reference Images are set. Default is direct.
seed
Optional integer for reproducibility. Leave blank for a random seed.
Use Cases
Film and episodic stills: Neo-noir, space opera, and horror frames with locked lighting language.
Product and beauty ads: Hero bottles, sneakers, and beverage splash shots for campaigns.
Game key art: Boss arenas, open-world vistas, and racing concepts without HUD clutter.
Localized posters and packaging: Multilingual headlines and calligraphy that stay legible.
UI and marketing mockups: Phone screens, landing sections, and app chrome with readable labels.
Reference-driven edits: Restyle a subject, relocate it, or composite 2 to 3 sources into one frame.
Tips for Better Results
Write a full paragraph. Subject, lighting, lens, and mood beat a short caption every time on these models.
Quote text you need to read. Put exact headlines and UI strings in quotes inside the Prompt.
Turn Prompt Extend off for finished prompts. It helps short prompts; it can dilute a shot you already locked.
Use Negative Prompt for clean film stills. Block text, logos, and UI when the goal is pure cinematography.
Order Reference Images deliberately. Name what to keep from image 1 versus image 2 in the Prompt.
Match Size to the deliverable. Widescreen film stills favor 2688×1536; posters and phone UI favor 1536×2688.
Use Pro for final frames. Iterate on Base, then regenerate the winner on Pro when fidelity matters.
Known Limitations
Prompt Extend Mode agent is text to image only. It is unavailable once Reference Images are attached; use direct instead.
Ultra-dense microcopy can still slip. Tiny body text in newspapers or menus may warp; keep critical strings large and quoted.
Reference order is the model's order. Swapping image 1 and image 2 can change which subject dominates.
Pro treats large Size presets as 2K. On Pro, sizes above about 1.5 megapixels are labeled 2K in the Size menu; Base keeps the same Size list without that split.