Black Forest Labs' open-weights image generation model with frontier performance, available for non-commercial local deployment
Settled by community votes across 7 shared challenges, with an AI judge weighing in on each.
FLUX.2 [dev]
#19 of 62 in Text-to-Image
Qwen Image Edit 2509
#32 of 32 in Image Editing
Where the votes landed
FLUX.2 [dev]
0.0%
win rate
Ties
0.0%
Qwen Image Edit 2509
100.0%
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
FLUX.2 [dev]
- + Excellent preservation of the man's identity, including his specific hairstyle, scarf, and plaid coat.
- + Faithful recreation of the car's interior details and ivory leather seats.
- + High-quality realistic lighting and motion blur on the road.
- − The camera angle focuses more on the driver than the car, cutting off most of the vehicle's exterior.
Qwen Image Edit 2509
- + Successfully merges the car and the man into a classic scenic California coastline shot.
- + Maintains the full exterior appearance of the Rolls Royce from the source image.
- + Provides a more dynamic 'driving' composition with motion on the wheels and road.
- − Loss of specificity in the man's appearance, including the removal of his scarf and patterned coat.
- − The man's face and hair are significantly lower in detail compared to the source.
Verdict: FLUX.2 [dev] is the winner because it successfully performs a complex identity transfer, keeping the man's specific clothes and features intact while placing him inside the car. Qwen Image Edit 2509 provides a better wide-angle composition of the requested scene, but it fails to preserve the unique fashion and facial details of the man from the source image.
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
FLUX.2 [dev]
- + Successfully integrated the man's head, glasses, and scarf from Image 2.
- + Maintained the yellow background and red ottoman from Image 1.
- − Nonsensical composition with two heads (man and woman's distorted face) in the same body.
- − Poor anatomical recreation with multiple limbs and floating elements.
- − The man's torso is upright, failing to match the leaning pose of Image 1.
Qwen Image Edit 2509
- + Maintains the original composition of Image 1 perfectly.
- − Completely failed to use the character reference from Image 2.
- − Basically returned a slightly color-processed version of Image 1.
- − Failed the instruction to change the identity of the person in the image.
Verdict: FLUX.2 [dev] attempted to follow the instructions by merging elements from both images, but the result is a surreal anatomical nightmare with two heads and disjointed limbs. Qwen Image Edit 2509 completely ignored the character reference in Image 2, simply regenerating the person from Image 1. Neither model was successful, but FLUX.2 [dev] at least attempted the edit, however poorly.
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
FLUX.2 [dev]
- + Successfully replicates the clothing style and accessories from Image 2.
- + Maintains high resolution and realistic fabric textures.
- + Good integration of the scarf and jewelry into the scene.
- − Failed to keep the person's exact face from Image 1, replacing it with a blend of both people.
- − Altered the person's skin condition/vitiligo pattern significantly.
- − Changed the hair length and style.
Qwen Image Edit 2509
- + Attempts to include the full outfit including the shoes requested.
- + Maintains the pose and leaning position of Image 1 better than the competitor.
- + Preserves the overall background context more accurately.
- − Failed to preserve the person's face from Image 1, modifying it into a new persona.
- − Low visual quality with significant artifacts and 'dirty' looking textures.
- − Very poor handling of the skin condition, making it look like face paint rather than the original vitiligo.
Verdict: Both models failed the primary constraint of keeping the person's exact face and hair from Image 1, instead opting to create a hybrid of the two subjects. FLUX.2 [dev] produced a much higher quality image with realistic textures and lighting, whereas Qwen Image Edit 2509 produced a low-fidelity result with heavy artifacts and a poor rendering of the subject's vitiligo. FLUX.2 [dev] is the winner for overall visual coherence and better adaptation of the clothing textures.
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
FLUX.2 [dev]
- + Excellent preservation of all facial features and original background details.
- + High-density hair that aligns with the 'full, thick' request.
- − The hair texture (afro) creates a mismatch with the existing straight beard texture.
- − The lighting on the hair feels slightly flat compared to the source image's harsh sunlight.
Qwen Image Edit 2509
- + Natural hair texture and color that matches the existing beard perfectly.
- + Highly realistic lighting and highlights that blend with the sunward side of the subject.
- + Accurate hairline and temple integration.
- − None notable; the edit is seamless and highly professional.
Verdict: Qwen Image Edit 2509 delivered a superior result by generating hair that perfectly matches the subject's existing beard texture and the scene's lighting, creating a seamless and realistic look. While FLUX.2 [dev] followed the prompt well, the choice of an afro texture felt less integrated with the subject's straight beard, making the edit appear more like a wig or an overlay.
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
FLUX.2 [dev]
- + Perfectly captures the caricature art style with exaggerated proportions and clean lines.
- + Successfully integrates all elements: anchor desk, hockey rink setting, and multiple dogs in hockey gear.
- + The facial features bear a strong, recognizable resemblance to the source image.
- − The text on the anchor desk is gibberish.
- − The man in the top left looks a bit disconnected from the rest of the scene.
Qwen Image Edit 2509
- + Maintains the 'selfie' pose and clothing (denim shirt) from the original source image.
- + The facial expression is very energetic and matches the 'humorous' instruction well.
- + Included a small dog and a hockey player in the background.
- − The dog element is very small and feels like an afterthought compared to the prompt.
- − The line art and coloring are a bit flatter and less polished than the first model.
- − Gibberish text in the speech bubble and background screens.
Verdict: FLUX.2 [dev] is the clear winner as it fully embraces the 'caricature' theme by creatively blending all requested elements into a cohesive scene, including dogs in jerseys on a hockey rink. While Qwen Image Edit 2509 preserves the original source's denim shirt and selfie pose better, it fails to integrate the dogs and hockey elements in a way that feels 'exaggerated and humorous' as requested.
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
FLUX.2 [dev]
- + Excellent adherence to the 'Studio Ghibli' aesthetic with soft, hand-painted textures and a watercolor feel.
- + Successfully captures the dreamy, nostalgic lighting and soft pastel color palette requested.
- + Maintains the composition and character poses while artistically elevating the background to a floral field.
- − Changed the background from a city street to a flower field, which deviates from the original context of the source image.
Qwen Image Edit 2509
- + Strongly preserves the source image composition, including the street, background characters, and bus.
- + Good transformation of faces into an anime style while maintaining clear emotional expressions.
- + Follows the instruction for pastel colors and soft lighting.
- − The texture feels more like digital cell shading/halftone than 'hand-painted textures'.
- − The line work on the man's hair and clothing feels a bit jagged compared to the soft Ghibli style.
Verdict: Both models do a great job at re-interpreting the 'Distracted Boyfriend' meme. FLUX.2 [dev] produces a much more authentic Studio Ghibli piece with its beautiful watercolor textures and atmosphere, although it replaces the urban background. Qwen Image Edit 2509 is better at source preservation by keeping the street setting intact, but its art style feels more like a generic web-comic than the specific Ghibli aesthetic requested.
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
FLUX.2 [dev]
- + Excellent preservation of the original face, clothing, and background details.
- + The added hair motion looks natural and maintains the original hair texture.
- + The flying leaves are realistically integrated with natural lighting.
- − The leash is slightly disconnected/glitched near the woman's hand compared to the original.
Qwen Image Edit 2509
- + Successfully added wind-blown hair and flying leaves.
- + Maintains the overall composition and color palette of the source image.
- − The hair has a 'liquified' or digital smudge look that lacks realistic fine stranded detail.
- − The leaves look like flat clip-art stickers rather than objects within the 3D space.
- − Subtle changes to the woman's facial features and teeth.
Verdict: FLUX.2 [dev] is the clear winner as it adds the requested motion while maintaining the high-fidelity realism of the original image, particularly regarding hair texture. Qwen Image Edit 2509 successfully applies the edits but introduces noticeable artifacts, making the hair look artificial and the leaves appear poorly composited over the scene.
Explore each model
Alibaba's Qwen image editing model for instruction-based image modifications and transformations