Quick Answer
Qwen-Image-Edit — Alibaba's open-source image editor with bilingual prompts. Install, ComfyUI workflow, when to pick Qwen vs Flux Kontext, and the niche where Qwen actually wins.
Quick Answer
Qwen-Image-Edit is Alibaba's open-source image-editing model — bilingual prompts (English + Chinese), 12GB VRAM, free commercial use up to 100M MAU. Install via ComfyUI Manager. Pick Qwen for text-in-image preservation and Chinese prompts; pick Flux Kontext for Western photoreal output.
What Qwen-Image-Edit Is
Alibaba released the Qwen-Image family in 2025 — Qwen-Image (text-to-image) and Qwen-Image-Edit (instruction-based editing). Qwen-Image-Edit is the editing variant: feed it a source image + text instruction, get the edited version.
- Bilingual prompts: Native English and Chinese support. Outperforms Flux Kontext on Chinese prompts.
- Text preservation: Best-in-class at preserving and editing text within images.
- VRAM efficient: Runs on 12GB GPUs (compared to Flux Kontext 16GB recommended).
- License: Tongyi Qianwen — commercial use up to 100M monthly active users (most creators won't hit this).
Install in ComfyUI
- Update ComfyUI to the latest version.
- Install ComfyUI Manager if you haven't — see our install guide.
- Install Qwen-Image custom nodes via Manager → Custom Nodes Manager → search “Qwen”. Click Install.
- Download model weights from Hugging Face:
Qwen/Qwen-Image-Edit— instruction-based editor (~10GB)Qwen/Qwen-Image— text-to-image base model (optional)
- Place weights in
ComfyUI/models/diffusion_models/ - Restart ComfyUI. Load the Qwen-Image-Edit example workflow from Templates → Image → Qwen.
ComfyUI Workflow
- Drop a source image into the Load Image node.
- Set the edit prompt in the conditioning node — keep it specific and short.
- Connect to Qwen-Image-Edit sampler with default settings (CFG 4-5, steps 30-50).
- Output goes to Save Image / Preview Image node.
- Queue prompt. ~30s on RTX 4090, ~90s on RTX 4070.
Prompt Patterns That Work
Qwen-Image-Edit responds well to instruction-style prompts in either English or Chinese:
- Outfit edit (English): “change the dress color to dark green, keep face and pose unchanged”
- Outfit edit (Chinese): “把裙子颜色改为深绿色，保持脸部和姿势不变”
- Background swap: “replace background with sunset beach, preserve subject lighting”
- Text edit: “change the sign text to ‘NEW STORE’ in same font and color” (Qwen excels here)
- Object addition: “add a glass of red wine in left hand, match scene lighting”
Qwen-Image-Edit vs Flux Kontext
| Aspect | Qwen-Image-Edit | Flux Kontext |
|---|---|---|
| Western photoreal | Strong | Best-in-class |
| Chinese prompts | Best-in-class | Limited |
| Text in image | Best-in-class | Strong |
| VRAM (12GB GPU) | Yes | Tight |
| Commercial license | Yes (up to 100M MAU) | Dev: non-commercial / Pro: paid |
| Model size | ~10GB | ~12GB (Dev), API only (Pro) |
| Speed (RTX 4090) | ~30s | ~30s |
Pick Qwen-Image-Edit when: you need Chinese prompts, edits to text within images, or want commercial-rights output without paying. Pick Flux Kontext when: you need the best Western photoreal portraits or your workflow already uses Flux Pro for hero generation.
Hardware Requirements
- Minimum: 12GB VRAM (RTX 3080 12GB, 4070 12GB)
- Recommended: 16GB+ VRAM (RTX 4080, 4090, A40)
- Cloud option: Runpod A40 ($0.40/hr) — 30s generation = $0.003 per image
Production Use Cases
- Bilingual brand campaigns — Western product photo edited with Chinese prompt for the China market.
- Text-heavy graphics — signs, posters, product labels with text changes preserving fonts.
- Outfit variations on a hero shot — generate one character image, run 20 outfit edits via Qwen for unlimited posts.
- Background variations — same character, different scene per post.
- Localization — translate text-in-image without re-shooting or re-designing.
Master Multi-Model AI Image Pipelines
Pros run multi-model stacks: Flux Pro for hero generation, Kontext or Qwen for variations, Wan 2.2 / Kling for animation. Our AI Influencers course teaches the full production pipeline scaling synthetic creators to $5K-$50K/month.
AI Influencers: The Multi-Model Stack
Flux + Qwen + Kontext + Wan 2.2 + Kling — full production pipeline for synthetic creators earning $5K-$50K/month.
Get AI Influencers →