Qwen Image 2.1 Prompts

Transparent assets · @PhotoX86

Use Qwen's recommended RGBA wording to create a cutout-style asset with an alpha channel instead of a baked background.

Transparent assets · @ItsmeAjayKV

Two high-resolution examples show the model producing transparent-background assets directly during local inference rather than relying on a separate background-removal pass.

Multi-reference · @junwatu

Assign one role to each reference image while preserving the room, camera and architecture as the fixed scene.

Multi-reference

A minimal official multi-reference instruction that gives three supplied characters one shared action and environment.

Multi-reference · @eternityspring

Use a single character image to anchor a widescreen sequence of domestic cooking beats with consistent mood, framing and understated performance.

Multi-reference · @PhotogenicWeekE

A ten-reference edit demonstrates dense subject placement and readable text while keeping the referenced characters recognizable in one composed result.

Multi-reference

Two concise editing prompts demonstrate product replacement under deformation and fitting a referenced watch while preserving the subject's pose, body shape, background and lighting.

Multi-reference · @yume_arasaki

A local workflow turns one character reference into expressions, turnarounds and action poses, then combines that sheet with a hardware photo in a text-heavy poster.

Multi-reference · @AIPixLab

The creator assigns one image to face and hair and front/side fashion photos to the garment, reporting a detailed rose-print transfer without training a clothing LoRA. No complete prompt is published.

Text & layout · @guitaripod

Stress-test exact labels, numbered instructions, a chart and physical props inside one realistic photograph.

Text & layout · @junwatu

Build a magazine-style profile poster with a preserved face, fixed title bar, information cards, swatches and a radar chart.

Text & layout

A compact official example that combines quoted display text with weather, surface reflections and a night setting.

Text & layout · @cocktailpeanut

Give each line of copy an explicit position while surrounding the sign with mundane industrial detail, fog and believable surface wear.

Text & layout · @cocktailpeanut

Combine a restrained packaging hierarchy with realistic aluminum, condensation and studio reflections in one commercial product shot.

Text & layout · @albertgao

Specify a legible two-letter mark, a tight palette and exclusions that keep a mobile app icon clear at small sizes.

Text & layout · @junwatu

Place one large portrait detail beside front, side and back full-body views while repeating the identity and costume constraints.

Text & layout · @cocktailpeanut

Combine an exact neon sign, a tiny roadside building and an enormous moon in a cinematic nighttime composition.

Text & layout · @sekiriously

A three-tier spot check compares typography, transparency, layout and fine detail at 1024 and native 2K while holding prompt, seed and step count fixed.

Precise edits · @VictorMotricala

Change hair, clothing and surroundings while explicitly protecting identity, facial features, skin tone and age.

Precise edits · @realrebelai

Use Qwen-Image 2.1 as its own generative upscaler, turning a 1024-pixel result into a larger image with a deliberately minimal edit prompt.

Precise edits · @SlipperyGem

Use a background-agnostic edit instruction to remove blur and improve fine detail while preserving layout, objects, colors, lighting, perspective and style.

Precise edits · @PhotogenicWeekE

A community comparison shows an existing cross-model sketch-conversion recipe transferring to Qwen Image 2.1, with visible seed-dependent differences in clothing and rendering.

Precise edits · @ajaeger74

A short edit instruction asks for higher resolution, restrained flash, finer feather detail and less noise without redesigning the subject.

Precise edits

The official single-image quick start shows that a direct background-change command can be enough for a basic edit.

Precise edits · @linoy_tsaban

Linoy Tsaban demonstrates an open-source relighting app built on Qwen Image 2.1. The video preview shows light-position controls; accelerated paths use Pruna or Viggle LoRAs rather than only the base model.

Creative generation · @ivanfioravanti

Describe impossible geometry as one continuous environment while locking the cinematic palette, materials and camera language.

Creative generation · @JiaMiXingTuan

Combine a concise style block with a specific industrial subject to produce a detailed widescreen science-fiction concept.

Creative generation · @cocktailpeanut

Make an ordinary late-night shop feel disturbing by combining flat fluorescent light, accidental framing and one subtly impossible figure.

Creative generation · @cocktailpeanut

Build a dreamlike scene around one visual substitution: flower petals rise from the engine where smoke would normally appear.

Creative generation · @Amrak101

Build a photorealistic horror frame from a frightened subject, a failing flashlight and a barely visible shape in the background.

Creative generation · @eternityspring

Plan a complete family outing, cooking sequence and moon-viewing finale as a coherent nine-panel illustrated storyboard.

Creative generation · @FiniYang

Use a restrained pose, directional window light and silver-grain texture to build a classic editorial portrait with a clean graphic silhouette.

Creative generation · @superalesha

A widely shared local-generation gallery shows Qwen Image 2.1 handling cinematic environments, unusual subjects and high-detail compositions without upscaling or retouching.

Creative generation · @ItsmeAjayKV

A 3840×2160 local result documents a demanding native-resolution run with the sampler, scheduler, step count, CFG and render time disclosed even though the rewritten prompt is not published.

Creative generation

A widely discussed gallery explores raw amateur phone-camera aesthetics through motion blur, low-light station noise and dynamic backgrounds.

Creative generation · @opener_ai

Four complete prompts replace polished AI aesthetics with self-timer mistakes, ultra-wide distortion, direct flash and uneven restaurant lighting.

Creative generation · @Brjen

An author-reported local RTX 3090 render uses a Qwen 3.8 Flash Next-written prompt to coordinate a stepped temple, terraced gardens, river and miniature depth of field.

Creative generation · @ericcurtin17

A short prompt from a local llmman demonstration. The shared image contains both a manatee and a sloth, so it illustrates the reported output rather than exact subject control.