ABrush · 17 Nov
How-to · 3 min read
How to edit an image with text prompts
Digital creators often struggle to modify specific elements in an existing image without altering the rest of the visual composition. When updating marketing assets, editorial illustrations, or digital product designs, visual creators frequently need to modify a single background element, alter an object color, or swap a subject outfit. Traditional generative artificial intelligence tools typically struggle with targeted revisions because generating a fresh image from an updated prompt renders an entirely new visual outcome from scratch. As a result, essential elements like existing subject poses, brand styling, ambient lighting, and camera perspectives are unintentionally changed or erased, wasting valuable creation time.
- Published
- 12 Nov 2026
- Delivery
- No. 11
- Topic
- Visual and media
- Evidence
- 9 claims quoted from qwenimage-21.org
01
Why it's hard
Editing images using standard text prompts is difficult because foundational visual generation models treat every generation pass as a completely independent, holistic process. When a prompt is adjusted to request a minor alteration, the model attempts to synthesize all visual elements simultaneously, which leads to unwanted shifts in background geometry, color temperature, and subject proportions. Relying on traditional graphic design software presents a different set of challenges, requiring tedious manual pixel selections, intricate magnetic lasso tracing, complex layer masks, and manual lighting adjustments. Without an integrated workflow that accepts explicit instructions defining both desired modifications and mandatory preservation areas, maintaining visual continuity across iterative revisions remains exceptionally hard.
This part describes the usual way. It makes no claims about qwenimage-21, so nothing is quoted here.
02
How qwenimage-21 does it
qwenimage-21 brings text-to-image creation and targeted image editing into a single unified workspace built for visual creators. The system enables users to describe visual edits clearly by identifying the exact object or region to modify while explicitly stating which background elements and framing constraints must remain unchanged. Creators can manage local reference drafts in the playground studio and select their preferred output aspect ratio before submitting an edit request directly to the model service. Every image modification or generation request follows a simple credit system based on chosen resolution, with failed generation requests being refunded automatically to the user account balance.
The model supports editing as well as creation. Name the area or object to change, then state what must stay.Quoted from qwenimage-21.org
For example: replace a grey sky with a warm sunset while preserving the building, camera angle and people.Quoted from qwenimage-21.org
Every image costs the same in generate and edit mode: 28 credits at 1K, 53 credits at 2K.Quoted from qwenimage-21.org
03
Step by step
- 1Access the playground studio and prepare your starting visual project or local reference image for an editing draft.
- 2Formulate a structured text prompt that explicitly names the targeted area, subject, or object you wish to replace or modify.
- 3Detail the key visual elements that must stay intact, including background details, lighting conditions, framing perspective, and color palettes.
- 4Choose the appropriate aspect ratio and image resolution settings, such as 1K or 2K, before submitting the request to the configured model.
- 5Inspect the generated outcome in the studio playground for text placement, edge boundaries, and subject details, refining prompt parameters one step at a time.
04
Tips
Structure your text prompt into two distinct halves: first describe the specific area or item to alter, and then explicitly list all elements, angles, and background features that must be preserved.
Name the area or object to change, then state what must stay.Quoted from qwenimage-21.org
Maintain brand consistency in commercial or product photographs by giving specific visual guidelines regarding exact object colors, packaging labels, and distinctive silhouette boundaries.
For a product image, mention color, label and silhouette when brand consistency matters.Quoted from qwenimage-21.org
Organize multi-reference workflows by assigning a clear, designated role to each reference image, such as defining the primary subject, outfit details, or background location.
Give each image a job: reference one for the subject, reference two for the outfit, reference three for the setting.Quoted from qwenimage-21.org
Review rendered outputs systematically for text accuracy and subject details, making adjustments to a single prompt constraint during each review pass.
Revise one constraint at a time.Quoted from qwenimage-21.org
05
Questions
How many credits does an image generation or edit cost?
Generating or editing an image costs 28 credits at 1K resolution and 53 credits at 2K resolution.
Every image costs the same in generate and edit mode: 28 credits at 1K, 53 credits at 2K.Quoted from qwenimage-21.org
Are credits refunded if an image generation fails?
Failed image generations are refunded automatically.
Failed images are refunded automatically.Quoted from qwenimage-21.org
Are the gallery illustrations outputs from user prompts?
No, the gallery visual studies are original site illustrations rather than outputs from user prompts or benchmarks.
These original studies are illustrations for the site. They are not outputs from your prompt or a benchmark of the model.Quoted from qwenimage-21.org
More on Visual and media
qwenimage-21 has one answer so far, so these come from its topic
PavoAI · 17 Nov
How to make a high-res image from a text prompt online
picttool · 17 Nov