Gemini 3.1 Flash with image generation capabilities. High-efficiency image generation model with support for text rendering, reference images, search grounding, and thinking mode. The efficient counterpart to Gemini 3 Pro Image.
Settled by community votes across 7 shared challenges, with an AI judge weighing in on each.
Nano Banana 2
#1 of 62 in Text-to-Image
Qwen Image Edit 2509
#32 of 32 in Image Editing
Where the votes landed
Nano Banana 2
0%
win rate
Ties
0%
Qwen Image Edit 2509
0%
win rate
Challenge by challenge
The strongest take from each model on every shared challenge, with the AI judge's read.
Man and Car in California
Editing“Make a photo of the man driving the car down the California coastline”
AI Judge Analysis
Nano Banana 2
- + Excellent preservation of the car's model and design features.
- + Highly accurate preservation of the man's facial features and specific hairstyle.
- + Convincing motion blur and lighting integration.
- − The man's scale relative to the car is slightly small, making him look a bit low in the seat.
Qwen Image Edit 2509
- + Beautiful composition and iconic California coastal scenery.
- + Good sense of speed and motion in the environment.
- − The man's facial features are largely lost to shadow and wind effects.
- − The hair texture is less detailed than in the source image.
- − Slightly less fidelity in the small details of the car's front grille compared to the source.
Verdict: Nano Banana 2 is the superior model for this task because it maintains remarkably high fidelity to the source images, specifically preserving the man's unique hairstyle and facial features perfectly. While Qwen Image Edit 2509 creates a beautiful scene, it loses much of the specific identity of the man in the process. Nano Banana 2 also does a better job of keeping the car's proportions and details consistent with the original photograph.
Pose & Character Mashup
Editing“Use Image 1 as the exact pose reference and Image 2 as the character reference. Recreate the person/character from Image 2 in the exact dynamic pose and body position from Image 1. Keep the exact face, hair, clothing style/details, and expression from Image 2. Match the lighting and environment of Image 1. The final image must show the character from Image 2 performing the precise action/pose from Image 1 with perfect anatomy and natural integration.”
AI Judge Analysis
Nano Banana 2
- + Successfully integrated the character, clothing, and accessories from Image 2 into the new scene.
- + Captured the pose and environmental lighting from Image 1 with high accuracy.
- + Maintained natural-looking skin textures and realistic fabric draping on the scarf.
- − The fingers on the raised hand appear slightly elongated and distorted.
- − The standing foot has an extra toe/anatomical blending issue with the red stool.
Qwen Image Edit 2509
- + Preserved the lighting and layout of the original scene perfectly.
- − Completely failed to incorporate the character reference from Image 2.
- − Only applied a minor color filter/saturation adjustment to the source image 1.
- − Failed to change the person's gender, clothing, or facial features as requested.
Verdict: Nano Banana 2 successfully followed the complex instructions to swap the character while maintaining the pose, clothing, and environment. Qwen Image Edit 2509 failed the task entirely, providing a slightly saturated version of the original pose reference without incorporating any elements from the character reference.
Outfit Transfer Challenge
Editing“Use Image 1 as the base person. Dress them in the exact elaborate outfit from Image 2 (including all layers, accessories, jewelry, and shoes). Carefully adapt the clothing to the body shape and pose in Image 1 while maintaining realistic fabric behavior, correct proportions, and perfect lighting/shadow matching. Keep the person’s exact face, hair, and background completely unchanged.”
AI Judge Analysis
Nano Banana 2
- + Successfully keeps the primary subject's head, hair, and vitiligo patterns mostly unchanged
- + Accurately recreates the textures of the coat and the specific pattern of the scarf
- + Maintains high resolution and realistic lighting consistency with the original beach scene
- − Hallucinates a second person in the background that was not requested or present
- − The hand and watch rendering on the foreground figure is slightly distorted
- − Does not include the black inner shirt from the reference outfit, leaving the chest exposed
Qwen Image Edit 2509
- + Successfully applies the full outfit including the under-layer black shirt
- + Positions the clothing to match a leaner body profile effectively
- + Maintains the background beach environment relatively well
- − Significantly alters the person's facial features and head shape, failing the source preservation requirement
- − Lower overall visual quality with heavy smoothing artifacts on face and clothing
- − Muddies the vitiligo patterns on the face, losing the distinct look of the original person
Verdict: Nano Banana 2 is the clear winner because it preserves the identity of the person in Image 1, which was a core instruction. While it hallucinates a second figure in the background, its rendering of the primary subject is much more realistic and respectful of the original features compared to Qwen Image Edit 2509, which completely changes the person's face and reduces image quality.
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
Nano Banana 2
- + Excellent source preservation of the face, glasses, and background
- + Extremely realistic hair texture and density
- + Perfectly natural-looking hairline and color match with beard
- − The hair volume is relatively conservative compared to Model B
Qwen Image Edit 2509
- + Successfully added a very 'full' and 'thick' head of hair as requested
- + Maintains original image elements like the jacket and background well
- − Hair looks like a wig rather than growing naturally from the scalp
- − The hair texture is overly stylized and wavy, appearing less realistic than the source textures
- − Distorted the facial features slightly, particularly making the forehead and glasses look slightly compressed
Verdict: Nano Banana 2 provided a much more plausible and photo-realistic result, with hair that perfectly matches the existing beard and lighting of the source image. While Qwen Image Edit 2509 followed the 'thick' instruction with more volume, the hair quality looks artificial and the integration with the subject's face is less seamless.
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
Nano Banana 2
- + Excellent adherence to all prompt elements including news anchor persona, multiple dogs, and hockey gear.
- + Strong text rendering with clever, relevant puns like 'W-K9 NEWS'.
- + Successfully captures the subject's likeness while applying a classic caricature style.
- − The 'paper-held-by-hand' presentation is a bit meta and wasn't specifically requested.
- − The hockey stick is being held in a somewhat awkward way by the hair/hand.
Qwen Image Edit 2509
- + Cleverly integrates the news desk directly onto a hockey rink.
- + Preserves the 'selfie' composition from the source image successfully.
- + Good caricature exaggeration of facial features.
- − The text in the speech bubble and on the screens is nonsensical/gibberish.
- − The dog is very small and lacks the 'love for dogs' emphasis requested compared to the hockey elements.
- − Introduced a secondary male character not mentioned in the prompt.
Verdict: Nano Banana 2 is the clear winner as it successfully incorporates all requested elements—hockey, dogs, and news anchoring—into a cohesive and humorous scene with high-quality, legible text. While Qwen Image Edit 2509 had a creative idea to place the news desk on a rink, it failed at rendering text and prioritized hockey over the request for dogs, resulting in a less balanced adherence to the prompt.
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
Nano Banana 2
- + Perfectly captures the Studio Ghibli watercolor aesthetic with hand-painted textures.
- + Excellently preserves the poses and the iconic 'distracted boyfriend' composition.
- + Enhances the background with Ghibli-esque floral details and charming architecture.
- − The woman in the foreground is slightly more blurry than the source, though it fits the depth of field.
Qwen Image Edit 2509
- + Good preservation of the source image's layout and character expressions.
- + Uses a soft, nostalgic color palette as requested.
- − Texture looks more like a digital screen tone or halftone dot pattern than hand-painted watercolor.
- − The lighting is somewhat flat compared to the requested dreamy Ghibli style.
- − The character in the foreground's face looks a bit generic and loses some of the source's likeness.
Verdict: Nano Banana 2 is the clear winner as it masterfully translates the source image into the specific Studio Ghibli art style, featuring beautiful watercolor textures and organic lines. Qwen Image Edit 2509 achieves a generic anime look with distracting digital patterns that fail to capture the 'hand-painted' requirement of the prompt.
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
Nano Banana 2
- + Successfully added energetic wind-blown hair and leaves that look natural to the scene.
- + Improved expression by changing the smile to a more joyous laugh that fits the mood.
- + Maintains high visual coherence with the dog's fur also reacting to the motion.
- − The facial features changed significantly from the source image.
- − Some small artifacts around the flying leaves.
Qwen Image Edit 2509
- + Preserved the subject's original facial features much better than Model A.
- + Successfully applied the hair motion and leaf additions requested.
- − The hair rendering looks somewhat artificial and 'stringy' compared to the source.
- − The added leaves look like flat clip-art stickers rather than integrated 3D objects.
Verdict: Nano Banana 2 provided a much more realistic and energetic interpretation of the edit, though it sacrificed the likeness of the woman in the process. Qwen Image Edit 2509 preserved the original face perfectly, but the hair and leaves look poorly composited and lack the 'dynamic feel' found in the other model's result.
Explore each model
Alibaba's Qwen image editing model for instruction-based image modifications and transformations