Alibaba Instruction-Based Image Editor
Alibaba's layout-aware editor takes direct commands — including exact copy changes in English, Chinese, or both — and reflows the image around your design instead of through it.
Your generated image will appear here.
Enter a prompt and click 'Generate' — or start from an example:
Examples
Drag the handle to compare each source and result, or open the viewer for a closer look.
text in image
Prompt: Change the header from "MORNING MENU" to "EVENING MENU" in the same decorative chalk capitals. In the left column, change "Matcha Latte 5" to "Ginger Tea 4.5"; in the right column, change "Seasonal Galette 6" to "Mushroom Toast 7". Keep the hand-lettered chalk style, the dark green board, the small illustrations, and the two-column layout exactly the same.
text in image, bilingual
Prompt: 把竖排的中文"茉莉花开"改成"冬日茉莉",保持书法字体、排版方向和位置不变。英文标题、盖碗、茉莉花和青瓷色调保持原样。 (Change the vertical Chinese text from "茉莉花开" to "冬日茉莉", keeping the calligraphy style, vertical layout, and position unchanged. Keep the English headline, the gaiwan, the blossoms, and the celadon palette as they are.)
semantic edits
Prompt: Change the oversized trench coat from crimson to deep emerald green satin with a subtle sheen, keeping the fabric's drape, the wind-blown movement, the folds, and the dramatic side light exactly the same. Keep the model, the concrete parking structure, the shadows, and the muted gray environment unchanged.
controlled changes
Prompt: Change step 2's text from "STEEP 12 hours" to "STEEP 16 hours" in the same type style, and change the small icon in step 2 from a clock to a crescent moon. Keep all other steps, numbers, cards, icons, connecting lines, colors, and the cold brew bottle illustration exactly the same.
text in image
Prompt: Change the curved text along the top of the ring from "CEDAR RIDGE" to "PINE HOLLOW" and the bottom text from "EST. 1974" to "EST. 1968", keeping the same distressed two-color screen-print lettering and the curve of each line. Keep the bear silhouette, the pine trees, the mountain peaks, the double-ring border, and the small stars exactly the same.
text in image + layout
Prompt: Change the masthead from "GRID" to "FRAME" in the same bold grotesque type, and change the cover line "THE NEW MINIMALISM" to "THE QUIET LAYOUT" in the same small aligned type. Keep the dancer photo, the yellow fabric, the concrete wall, the other cover lines, and the overall grid and hierarchy exactly the same.
Qwen Image Edit descends from Alibaba's Qwen Image generator, a model known for rendering real text, and the editing version inherits that literacy: it can change the words inside an image — in English, in Chinese, or both at once — while respecting the layout around them. Signs, menus, slides, and storefronts keep their design logic after the copy changes.
Beyond text, it performs semantic edits — swapping objects, shifting styles, adjusting scenes — from direct instructions rather than masks. On RenderFlow AI it runs from a single uploaded image at 3 credits per edit, with JPEG, PNG, and WebP export, making it the most affordable editor in the image-edit lineup.
The canvas takes a single input image, so choose the version with the right composition and design bones. Qwen Image Edit preserves what your instruction does not name — there is no masking step to do first.
Phrase the edit as an order — "change the sign to read '营业中 OPEN'" or "swap the red mug for a glass teapot" — and quote any text verbatim, in either language or both. Direct single-action commands are what the model follows most reliably.
Say where the change lives — top-left, second line, behind the counter — and name the elements that must stay untouched. Layout awareness is the model's strength, but it needs the instruction to acknowledge the design it is preserving.
JPEG suits photographic results, PNG suits graphics and anything headed into a design file, and WebP suits web delivery. Once a direction looks right, pin the Seed so retries refine instead of restart.
When the job needs reference images, precise output dimensions, or tougher packaging-grade text, move the same instruction to Qwen Image Edit Plus at 4 credits. Single-image copy fixes belong here.
Yes — bilingual Chinese and English text editing is its signature capability, inherited from the Qwen Image generator. Quote the exact characters you want rendered, and the model resets them in place while the surrounding layout stays intact.
Direct, single-action commands: change this copy, swap this object, restyle this scene. It handles semantic edits and text edits from plain sentences, and position words like "top-left" or "second line" noticeably improve precision.
The standard model edits one image at 3 credits with format and seed controls. Edit Plus is the 20B dual-mode tier at 4 credits: it accepts up to three input images, sets exact output dimensions from 256 to 1536 pixels, and delivers stronger precision on demanding text and packaging work.
JPEG, PNG, and WebP are all selectable in the canvas. JPEG fits photographic edits, PNG fits graphics and design handoff, and WebP keeps web-delivered assets light.
The instruction usually failed to name what must stay. Add explicit anchors — "keep the logo bar and footer untouched, change only the headline" — and the model reflows around the preserved elements instead of redesigning them.
For text and layout-adjacent edits, yes: at 3 credits per image it is the cheapest editing model on the platform, and its bilingual text precision covers most everyday copy changes before you consider a premium tier.
Create images with strong text rendering and layout control.
Use advanced Alibaba Qwen image editing for product and layout changes.
Edit existing images with Google Nano Banana Edit.
Edit images with context-aware FLUX prompt controls.
Create controllable 720p or 1080p AI videos from text prompts with Alibaba Wan 2.7.
Animate first and optional last frames at 720p or 1080p with Alibaba Wan 2.7.