Alibaba Production-Grade Image Editor
The 20B bilingual editing tier for production work — label swaps, layout-precise changes, and exact-size output for the slot the asset has to fill.
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.
object merge
Prompt: Apply the vintage label design from the second image onto the candle jar in the first image, wrapping it naturally around the jar's curve with correct perspective, soft paper texture, and the studio lighting falling consistently across it. Keep the jar's shape, amber glass, lid, and background exactly the same, and keep all label text and artwork readable.
Source: Double--M — CC-BY 2.0 via Flickr.
object merge, identity preserved
Prompt: Place the woman from the first image sitting at one of the rattan tables in front of the café in the second image, as if mid-conversation over a coffee. Match the daylight direction and color temperature of the street scene, add a natural soft shadow at her chair, and keep her face, hairstyle, and blazer exactly as in the original portrait so she is unmistakably the same person.
Source: decar66 — CC-BY 2.0 via Flickr.
object merge
Prompt: Place the transparent pebble speaker from the first image on a rain-wet stone ledge in the night street scene from the second image. Match the ambient neon lighting, add colored reflections on the wet stone around the speaker, and keep the speaker's shape, transparency, internal drivers, and LED ring exactly as in the product shot.
Source: Steven Pisano — CC-BY 2.0 via Flickr.
source preservation
Prompt: Repaint the overhead lunch photo from the first image in the loose watercolor style of the second image: visible paper texture, soft pigment bleeds, hand-painted edges, the same muted blue-and-ochre tonality. Preserve the exact composition of the first image — toast with burrata centered, tomatoes and basil right, carafe left, napkin and fork at the edge.
Source: rawpixel — CC-BY 2.0 via Flickr.
object merge
Prompt: Print the national park badge from the second image onto the front of the tote bag in the first image, centered, following the fabric's slight wrinkles and folds with a realistic screen-print texture. Keep the bag's shape, canvas color, straps, wall, and lighting exactly the same, and keep the badge artwork and text crisp.
multi-image edit, face preserved
Prompt: Change the man's navy sweater in the first image to a shirt made from the green plaid fabric in the second image, with the plaid pattern following the folds and seams naturally. Keep his face, gray hair, stubble, pose, expression, and the soft gray studio background exactly the same.
Source: webtreats — CC-BY 2.0 via Flickr.
Qwen Image Edit Plus is Alibaba's 20B-parameter MMDiT editing model — bilingual in Chinese and English, and dual-mode, meaning one instruction can combine appearance changes (color, material, finish) with semantic ones (replace this object, swap this label). Style preservation is explicit: brand elements that are not named in the prompt carry over unchanged.
On RenderFlow AI, the canvas accepts up to three input images — a packshot plus label flats, logos, or lifestyle references — and renders at exact width and height settings from 256 to 1536 pixels. At 4 credits per edit with JPEG, PNG, and WebP export, it sits between the everyday editors and the premium Kontext tier.
Upload the product or layout image as Input Image 1, then enable Inputs 2 and 3 for the material the edit should incorporate — the new label flat, the logo lockup, a lifestyle reference. Concrete artwork in the inputs beats describing artwork in the prompt.
State the surface, the copy, and the finish in one instruction: "replace the front label with image 2, keep the foil stamp and the bottle shape." Quote label strings verbatim in whichever language they print — the model is bilingual by design.
Both dimensions adjust independently from 256 to 1536 pixels (1024×1024 default). Match the marketplace image slot or banner placement exactly so the deliverable needs no resizing after render.
Choose PNG for design handoff, JPEG for listings, WebP for web delivery. Pinning the Seed once a result is approved lets you rerun the same brief across a SKU line with matched treatment.
A single-image copy tweak with no size spec does not need the Plus tier — the standard model does it for 3 credits. Plus earns its keep when multiple inputs, exact dimensions, or packaging-grade text are in play.
The model handles appearance-level changes (color, texture, material, finish) and semantic changes (replacing objects, swapping labels, altering scene content) within a single instruction, while preserving the style of everything the instruction does not name.
Yes — the 20B model is bilingual by design. Quote each string verbatim in its language and describe the treatment ("same condensed type, white on black"), and the new copy renders into the packshot rather than overlaying it.
Width and height are set independently anywhere from 256 to 1536 pixels, with 1024×1024 as the default. That covers marketplace slots, listing images, and social placements at exact spec.
Up to three: one required base image plus two optional references. Typical production use is a packshot as the base with the label flat and logo lockup as references.
Upgrade when the job needs more than one input image, exact output dimensions, or text and layout precision that must survive scrutiny — packaging, key art, branded templates. Single-image copy fixes on flexible sizes are fine on the 3-credit standard model.
Type below a renderable size in the frame is a hard case for every current image editor. Crop tighter on the text region, shorten the string, or enlarge the text area in the output size; headline-scale copy is where the model is reliable.
Edit images with Alibaba Qwen prompt-based controls.
Create images with strong text rendering and layout control.
Use the premium FLUX Kontext image editing model.
Edit images with ByteDance Seedream V4 Edit.
Create higher-quality MiniMax text-to-video generations.
Animate images with premium MiniMax Hailuo video generation.