Tongyi Qwen · Qwen-Image-Edit · qwen-image-edit
Tongyi Qwen's instruction image editing tier. Edit CN/EN text on existing images, add/remove objects, outfit and IP derivatives—while preserving font style and non-target areas as much as possible.
Add/remove objects, outfit and background swaps are appearance editing; IP derivatives, rotate viewpoint, style transfer are semantic editing. Client wants one change but you regenerate whole frame—deadline gone.
Graffiti wall Hope→Qwen, Chinese poster headline swaps—all editable; text right but style drifted still fails review. Specify what font and decoration to preserve in prompts.
Official demos include Chinese poster fine-type edits and calligraphy wrong-character step-by-step box fixes—miss one rare character, whole calligraphy piece undeliverable.
Appearance editing emphasizes non-target area preservation; identity drift kills whole series hero images. With multi-image input, name "image 1/image 2" explicitly to avoid wrong asset reference.
Official / Bailian demos: text add/remove, local recolor, IP multi-look, pattern on merch, figurine, multi-image character compositing—check "what changes, what stays."
Vs Plus, Wan 2.7, GPT Image 2. Editing default Edit; from-scratch long copy Plus; ~2K generate+edit Wan 2.7; OpenAI stack GPT Image 2.
| Dimension | Qwen Image Edit | Qwen Image Plus | Wan 2.7 | GPT Image 2 |
|---|---|---|---|---|
| Editing | Instruction editing · add/change/delete text | Mainly text-to-image (no editing this tier) | Instruction editing · add/change/delete text | Instruction editing · add/change/delete text |
| In-image text | Short copy readable · text editable | Complex CN/EN long copy (text-to-image) | Short copy readable · text editable | Short copy readable · text editable |
| Local precision | Short copy readable · text editable | Short copy usable | Short copy readable · text editable | Short copy readable · text editable |
| Resolution | Size not specifiable · defaults ~follows input (≈1K) | Fixed presets, ~1K–1.5K class (five tiers) | Up to ~2K | Native 4K (3840×2160) |
| Chinese support | Strong Chinese text support | Strong Chinese text support | Strong Chinese layout · editable | Multi-reference support |
| Reference / input images | 1–3 input images | No image input (text-to-image only) | Multi-reference support | Multi-reference support |
| Speed profile | Quality-first · not instant draft | Quality-first · not instant draft | Quality-first · not instant draft | Faster output |
| Default pick | Precise editing & CN text change default | Chinese posters & long copy first | Wan generate+edit default (standard tier) | OpenAI new-project default |
Text changes, add/remove, identity stays put, Chinese posters fixable—these differentiate it from pure text-to-image channels.
Natural language specifies what to change and keep; full regeneration loses locked lighting and composition.
Graffiti, posters, sign text changes; drifted font style still makes brand mock unusable.
Chinese poster small type and wrong-character chain box fixes; miss one character, whole piece undeliverable.
Add sign/remove clutter/swap hat; non-target area drift doubles retouch rounds.
Ghibli-style conversion, rotate to back view, IP multi-look; semantic drift breaks character series alignment.
1–3 inputs for outfit swap, pose transfer; wrong image index reference, wrong subject in result.
Mapped to this page's samples: text changes, local add/remove, IP derivatives, pattern on merch, figurine productization, multi-image character fusion—not generic "make it prettier."
Graffiti Hope→Qwen, poster headline swap; text right but style drifted still fails review.
Remove car keep garage, petals to blue, add top hat; rest of frame changes too—equals redo.
Same mascot derivative career looks; face and brand color drift kills whole sticker pack.
Print illustration on T-shirt and paper bag with model; wrong perspective and folds make e-commerce hero untrustworthy.
Character to figurine, packaging and screen modeling comparison; wrong scale fails pitch review.
Character compositing + scene base fusion; wrong "image 1/image 2" reference, wrong subject placement.
Default to Edit when: you already have review-ready original, only changing text, props, or local area—failure cost is locked frame and schedule.
Hero clutter removal, background swap, price tag text; full regen every SKU eats schedule.
Change text and props on locked composition, then decide Plus re-gen or other polish channel.
Campaign visual headline changes, localization text swap; drifted font style kills ad materials.
IP multi-look, comic character swap, sticker series; identity drift breaks update rhythm.
Fewer full redos when client changes one line or prop; side-by-side Plus / Wan / GPT comparison.
Clutter cleanup, local recolor, outfit trials; skin and light fully drifted equals reshoot.
Select Qwen Image Edit (qwen-image-edit) in iMini's image workspace—same account as Plus, Wan, GPT Image.
Upload 1–3 images; specify what to change and keep; multi-image use "image 1/image 2" labels; quote text to change.
Verify keywords and identity; box-select local areas for more rounds if needed. From-scratch long-copy posters—switch to Plus.
Matched to this page's capabilities. Hover for full text, one-click copy; always specify preserve areas and text to change.
Preserve wall brick texture, lighting, and vine decoration style, only change graffiti text from "Hope" to "Qwen", match original graffiti stroke weight, do not change surrounding plants or wall stains.
Preserve face, clothing, and background unchanged, only replace graduation cap with purple Victorian top hat, natural brim-to-head ratio, do not change expression or pose.
Keep this Qwen bear's face shape, colors, and T-shirt logo unchanged, change to business look in suit carrying briefcase, still signature dot eyes, white background, flat illustration style.
Print this guitar-playing bear pattern on white T-shirt and white paper tote bag, female model holding tote facing camera, black baseball cap reading "Be Kind", clothing store background blurred, keep pattern clear and readable.
Before generating: free quota, vs Plus, how many input images, resolution, text-change positioning.
Instruction text changes · add/remove objects · 1–3 inputs—compare side-by-side with Plus, Wan 2.7, GPT Image 2.
Generate free on iMiniNo vendor API key—iMini account is enough.
