Alibaba's Qwen image editing model for instruction-based image modifications and transformations
Settled by community votes across 4 shared challenges, with an AI judge weighing in on each.
Qwen Image Edit 2509
#32 of 32 in Image Editing
Z-Image Turbo
#12 of 62 in Text-to-Image
Where the votes landed
Qwen Image Edit 2509
0%
win rate
Ties
0%
Z-Image Turbo
0%
win rate
Challenge by challenge
The strongest take from each model on every shared challenge, with the AI judge's read.
Bald man challenge
Image Editing“Give the person a full, thick head of natural hair with realistic texture, density, and a natural hairline. Preserve facial features and lighting.”
AI Judge Analysis
Qwen Image Edit 2509
- + Successfully added a thick head of hair while preserving the subject's face and original setting perfectly.
- + Modern, high-quality hair texture that blends reasonably well with the existing beard.
- + Maintains the exact glasses, facial expression, and lighting from the source.
- − The hairline on the forehead looks slightly superimposed and lacks fine transitional hairs.
- − The hair style is a bit overly dramatic or styled compared to the source's rugged aesthetic.
Z-Image Turbo
- + Maintains the overall composition and lighting of the original scene.
- − Failed to add a full, thick head of hair as requested, only adding a very short buzz cut or stubble.
- − Removed the subject's glasses without being asked.
- − Slightly altered the facial features and altered the background scenery.
Verdict: Qwen Image Edit 2509 successfully completed the core task by giving the subject a full head of hair while keeping the identity and glasses intact. Z-Image Turbo failed twice: first by only adding a micro-buzz cut instead of thick hair, and second by removing the subject's glasses and altering the background.
Over-the-top cartoon caricature
Editing“Create a caricature of me and my job. Make it exaggerated and humorous, incorporating my profession as a tv show anchor and my love for dogs and hockey.”
AI Judge Analysis
Qwen Image Edit 2509
- + Successfully incorporates all prompt elements: TV anchor desk, hockey rink background, and a dog.
- + Captures the 'caricature' style with exaggerated facial features and a comic-book aesthetic.
- + Preserves the subject's clothing and pose while adapting them into the new style.
- − The text in the speech bubbles and monitors is nonsensical.
- − The dog at the desk is very small and lacks detail.
Z-Image Turbo
- + Excellent source preservation, keeping the subject's face almost identical to the original.
- + Subtly adds a small dog in the background.
- − Fails to follow almost all instructions: no caricature style, no TV anchor setting, and no hockey elements.
- − The image is essentially just the source image with minor lighting adjustments and a blurry dog added.
Verdict: Qwen Image Edit 2509 is the clear winner as it fully embraced the complex prompt, transforming the image into a humorous caricature featuring a news desk, a hockey rink, and a dog. Z-Image Turbo almost entirely failed the edit request, providing a realistic portrait that ignored the caricature style and the hockey/TV anchor themes.
Studio Ghibli Anime Style
Editing“Transform this photo into a Studio Ghibli–inspired illustration. Use soft pastel colors, hand-painted textures, gentle lighting, dreamy backgrounds, and a warm, nostalgic mood”
AI Judge Analysis
Qwen Image Edit 2509
- + Excellent adherence to the 'Studio Ghibli' style, using clean line art and hand-painted textures.
- + Preserves the iconic composition and poses of the original 'distracted boyfriend' meme perfectly.
- + Applied the requested pastel color palette and warm, nostalgic lighting effectively.
- − The transition of the red dress to a light pink changes the fundamental color identity of the original image.
- − The background characters in the distant center are simplified into generic shapes.
Z-Image Turbo
- + Successfully preserved all original colors and subject details.
- + Maintains the photographic quality of the source image.
- − Failed the edit instruction completely by not transforming the image into an illustration style.
- − Zero detectable influence from Studio Ghibli or 'hand-painted textures'.
- − The person on the right has their facial expression neutralized compared to the indignant look in the original.
Verdict: Qwen Image Edit 2509 is the clear winner as it successfully interpreted and applied the stylistic transformation requested, creating a cohesive anime-style illustration that respects the original composition. Z-Image Turbo failed to apply the requested artistic style, essentially outputting a slightly altered version of the original photo without any illustrative qualities.
Golden Hour Stroll
Image Editing“Add dynamic motion to this photo: make hair blow in the wind, add leaves flying, energetic and lively feel.”
AI Judge Analysis
Qwen Image Edit 2509
- + Strong presence of flying leaves that adds significant visual energy.
- + Expressive hair movement that clearly responds to the wind request.
- + Preserves the subject's face and the dog's appearance almost perfectly.
- − The hair strands appear somewhat repetitive and artificial in their waves.
- − The leaves look a bit like stickers layered on top rather than being integrated into the scene's lighting.
Z-Image Turbo
- + Natural and subtle hair movement that looks physically realistic.
- + Better integration of the few leaves added into the environment's lighting.
- − Fails to preserve the subject's identity, completely changing the woman's facial features.
- − The number of leaves is very small, missing the 'energetic and lively' part of the prompt.
- − The composition and background details (flowers, bridge area) were unnecessarily altered/smudged.
Verdict: Qwen Image Edit 2509 is the clear winner as it successfully follows the prompt's instructions for dynamic motion and flying leaves while maintaining the identity of the person in the source image. Z-Image Turbo failed as an image editor by completely changing the woman's face and only adding a few leaves, resulting in a less energetic feel.
Explore each model
Tongyi-MAI's 6-billion parameter distilled text-to-image model optimized for speed, achieving high-quality generation in 8 steps or fewer with support for bilingual text rendering