AI热点 3小时前 62 阅读 0 评论

Flux.1 Krea Dev超大杯实测:开源模型能否撼动Midjourney V7 ?

作者头像
AI中国

AI技术专栏作家 | 发布了 246 篇文章

嗨大家好!我是阿真!


7月底 Black Forest Labs 和 Krea 合作开发的高级文本到图像生成模型 Flux.1 Krea Dev,最近终于有时间进行测评了。


Flux.1 Krea Dev 是基于FLUX.1 dev 模型进行蒸馏的,参数规模12B,专注于提升图像的美学和真实感,避免了常见的 AI 生成痕迹(过度饱和或不自然高光等等),更倾向于追求自然细节、照片级真实感和多样性。


模型的核心目标是通过少量高质量数据集和自定义训练技术来实现“有主观色彩的(opinionated)”风格,强调多样性、自然细节和照片级真实感。它在内部评估中优于之前的开源 FLUX 模型,并接近闭源模型如 FLUX.1 Pro 的性能,尤其在提示遵循、视觉质量和多样性方面


看网友评论说这是目前最好的开源 FLUX 模型,我的好奇心就上来了。


今天一起来看看它是不是真的那么优秀。


目录


1. 人物与动物


2. 场景


3. 特定风格


4. 其他


5. 小结


先下载 Flux.1 Krea Dev 。


ComfyUI官方流程:https://docs.comfy.org/tutorials/flux/flux1-krea-dev


我是直接按照上方链接中官方的教程从ComfyUI官方客户端这里下载的。



其实文生图模型的竞争早已进入白热化阶段,一般的细节上的提升已经难以被普通人用肉眼察觉了。


今天,我将拉上(我认为的)目前最强的闭源文生图模型 MidJourney V7,与它进行对比,让大家一探究竟。不是比谁更强,就是看看还有多远的距离。


不过这个对比完全不严谨,仅用于给大家进行对比大致的图片效果。因为图片细节在这里会被压缩,另外大家上手体验后个人的体验感都会有差异。设备局限,我今天使用的是低精度的t5xxl_fp8_e4m3fn.safetensors,高精度的效果相比可能还会更好一些。


我今天用的电脑显卡是4080s,内存32G,生成一张1360*784的图片大约需要20秒。



那么直接开始吧!


请注意,下列图片,两张对比图的时候,上图为Flux.1 Krea Dev生成,下图为Midjourney V7生成。


1.人物与动物


这一轮我们主要测试人物表情、服饰与环境的自然整合。




上图-Flux.1 Krea Dev / 下图-Midjourney V7


A joyful child holding a colorful toy, playing in a sunny playground, dynamic low-angle shot, soft sunlight casting playful shadows, pastel colors, intricate details, 8k resolution


一个快乐的孩子拿着彩色玩具,在阳光明媚的游乐场玩耍,动态低角度拍摄,柔和的阳光投射出俏皮的阴影,柔和色彩,精致细节,8k分辨率。




上图-Flux.1 Krea Dev / 下图-Midjourney V7


Model standing in front of red velvet curtains, wearing black fringe evening gown, holding silver clutch bag, dramatic lighting, ad vibe


模特站在红色天鹅绒窗帘前,穿着黑色流苏晚礼服,手持银色手拿包,戏剧性灯光,广告氛围


这里可以看到全景镜头和远景镜头上Flux.1 Krea Dev的细节就有点跟不上了。可能提示词详细程度也有影响,看看详细提示词的效果↓




上图-Flux.1 Krea Dev / 下图-Midjourney V7


A rugged bearded man hiking across a vast snow-covered mountain range, photorealistic, soft diffused sunlight peeking through the clouds, gently illuminating the glistening snow and jagged peaks in the background, intricate textures of his windproof jacket, each wrinkle and seam of the clothing is visible, backpack straps tightly secured with small reflective patches catching the light, detailed stitching and fabric folds on his durable boots, snowflakes gently drifting around him, the cold, crisp air visible in the atmosphere, distant rocky ridges and frozen lakes dotting the landscape, dynamic camera angle that captures his determined expression while focusing on the tactile realism of the environment, hyperdetailed, 8k resolution, sharp textures with visible grain, each snowflake"s crystal structure rendered in high detail, advanced lighting and shadows for a cinematic effect, filmic grain for a natural, earthy feel.

(太长了省略翻译了)


我个人认为Flux.1 Krea Dev 使用更详细的提示词可能会得到更好的效果,Midjourney目前来说简单或详细的提示词对于图片精度细节的影响已经不大了。




上图-Flux.1 Krea Dev / 下图-Midjourney V7


A stunning supermodel draped in an opulent, orange-toned gown, wearing exquisite jewelry, standing gracefully in a luxurious, golden-lit room, full body shot, photorealistic, dramatic warm lighting that highlights the sparkling gemstones and the flowing fabric of her gown, intricate detailing in the gold and diamond accessories, the soft texture of the silk dress with its elegant folds, soft focus on her serene and confident expression, the background filled with rich, ornate patterns and soft ambient light, giving a sophisticated and glamorous atmosphere, hyperdetailed, 8k resolution, sharp focus on textures, refined lighting casting soft shadows that enhance the luxurious mood of the scene


一位绝美超模身披华丽的橙色调长裙,佩戴精美珠宝,优雅地站在奢华的金色灯光房间中,全身镜头,照片级写实,戏剧性的暖色调灯光突出了闪闪发光的宝石和长裙飘逸的面料,黄金和钻石配饰的精致细节,丝绸裙装柔软的质感及其优雅的褶皱,她宁静自信的表情采用柔焦处理,背景充满丰富华丽的图案和柔和的环境光,营造出精致迷人的氛围,超精细,8k分辨率,纹理锐利聚焦,精致的灯光投射出柔和阴影,增强了场景的奢华氛围。




上图-Flux.1 Krea Dev / 下图-Midjourney V7


An Asian woman standing confidently in front of a towering mountain peak, her hair blowing in the wind, full body shot, photorealistic, soft golden hour light illuminating her face and the rugged texture of the mountain, the wind tousling her hair with delicate strands flying in the air, mid-shot composition, eye-level angle, serene and determined expression, vibrant color palette with natural greens and cool mountain grays, shot with an 85mm lens at f/2.8, shallow depth of field creating a soft background blur, hyperdetailed, 8k resolution, the textures of her clothing gently swaying in the breeze, capturing the vastness of the natural landscape


一位亚洲女性自信地站在高耸山峰前,头发在风中飞舞,全身镜头,照片级写实,柔和的黄金时刻光线照亮她的面庞和山峰粗糙的质感,微风吹拂着她的头发,精致的发丝在空中飞舞,中景构图,平视角度,宁静而坚定的表情,鲜艳的色彩调板搭配自然绿色和清冷的山峰灰色,使用85mm镜头f/2.8光圈拍摄,浅景深创造出柔和的背景模糊,超精细,8k分辨率,她的衣物质感在微风中轻柔摆动,捕捉自然景观的广阔壮丽。




上图-Flux.1 Krea Dev / 下图-Midjourney V7


An elderly man with glasses, sitting comfortably in a cozy armchair, reading a book with intense focus, close-up shot, photorealistic, soft ambient light from a nearby lamp casting a warm glow on his face and the pages of the book, intricate details in the textures of his worn-out sweater and the leather-bound book, the pages slightly yellowed with age, serene atmosphere, soft shadows, rich wood tones of the bookshelf behind him, hyperdetailed, 8k resolution, the wrinkles on his face and the calm expression reflecting years of wisdom.


一位戴着眼镜的老年男性舒适地坐在温馨的扶手椅中,专注地阅读一本书,特写镜头,照片级写实,附近台灯发出的柔和环境光在他的面庞和书页上投射出温暖的光辉,他破旧毛衣和皮装书籍的质感细节精致入微,书页因岁月而略显泛黄,宁静的氛围,柔和的阴影,身后书架丰富的木质色调,超精细,8k分辨率,他脸上的皱纹和平静的表情反映出岁月积淀的智慧。




上图-Flux.1 Krea Dev / 下图-Midjourney V7


Elderly woman and giant cat standing in the center of a lawn, background residential buildings symmetrically aligned, clear depth layers, central symmetry composition, natural light, balanced colors, realistic style, 35mm lens, high resolution.


老年女性和巨型猫咪站在草坪中央,背景住宅建筑对称排列,清晰的景深层次,中心对称构图,自然光线,色彩平衡,写实风格,35mm镜头,高分辨率。




上图-Flux.1 Krea Dev / 下图-Midjourney V7


A warrior stands amidst the burning ruins of a battlefield, firelight reflecting off the metallic textures of his armor, blood dripping from his longsword, surrounded by fallen enemies and shattered weapons, background a mix of smoke and flames, 50mm lens, f/2.8, high dynamic range lighting, strong warm-cool contrast, razor-sharp details.


一位战士站在燃烧的战场废墟中,火光在他盔甲的金属质感上反射,鲜血从他的长剑上滴落,周围散布着倒下的敌人和破碎的武器,背景是烟雾和火焰的混合,50mm镜头,f/2.8光圈,高动态范围照明,强烈的暖冷对比,锐利清晰的细节。




上图-Flux.1 Krea Dev / 下图-Midjourney V7


A happy couple in mid-close-up, holding hands and laughing joyfully, their faces radiating happiness under the soft golden light of sunset, photorealistic, soft sunlight casting gentle highlights on their skin, surrounded by lush greenery and vibrant flowers, the background slightly blurred to emphasize their expressions and bond, natural light creating a warm and intimate atmosphere, hyperdetailed, 8k resolution, every delicate detail of their facial expressions, clothes, and the surrounding nature captured with clarity and warmth


一对幸福的情侣中近景镜头,手牵手欢声笑语,他们的面庞在日落柔和的金色光线下散发着幸福,照片级写实,柔和的阳光在他们的肌肤上投射出温柔的高光,周围环绕着茂盛的绿植和鲜艳的花朵,背景略微模糊以突出他们的表情和情感纽带,自然光营造出温暖而亲密的氛围,超精细,8k分辨率,他们面部表情、服装以及周围自然环境的每一个精致细节都以清晰和温暖的方式捕捉




上图-Flux.1 Krea Dev / 下图-Midjourney V7


Two young people standing in modern art gallery hall, half-body shot, left in black fitted blazer with gray silk shirt, right in light blue blazer with white knit turtleneck, minimalist marble wall and art installation in background


两个年轻人站在现代艺术画廊大厅里,半身镜头,左边穿着黑色修身西装外套配灰色丝质衬衫,右边穿着浅蓝色西装外套配白色针织高领毛衣,背景是极简主义大理石墙面和艺术装置




上图-Flux.1 Krea Dev / 下图-Midjourney V7


Black and white photography, man leaning against white wall in a minimalist gallery, wearing white shirt and high-waisted dress pants, sleeves rolled up, blurred abstract painting and floor-to-ceiling window in background, soft natural light


黑白摄影,男子靠在极简主义画廊的白墙上,穿着白色衬衫和高腰西装裤,袖子卷起,背景是模糊的抽象画和落地窗,柔和的自然光线




上图-Flux.1 Krea Dev / 下图-Midjourney V7


A cute young girl with short gray-blue hair, wearing a futuristic mech battle suit with a sleek white and black design, the chest adorned with orange and black details, full body shot, photorealistic, dynamic pose with her hands forming a V-sign, the suit"s intricate design highlighting its mechanical components, the contrasting colors of white, black, and orange giving it a bold and striking look, soft ambient lighting casting shadows that emphasize the contours of her suit and facial features, vibrant color palette with cool tones and a slight futuristic glow, hyperdetailed, 8k resolution, the suit’s metallic texture and the smoothness of her skin captured with clarity, adding a sense of strength and youthful charm


一个可爱的短发灰蓝色头发少女,身穿未来主义机甲战斗服,设计流畅,以白色和黑色为主,胸部装饰着橙色和黑色细节,全身镜头,照片写实风格,动态姿势双手做出V字手势,战斗服的精细设计突出其机械组件,白色、黑色和橙色的对比色彩营造出大胆醒目的外观,柔和的环境光线投下阴影,强调她战斗服和面部特征的轮廓,鲜艳的色彩调色板以冷色调为主,带有轻微的未来主义光辉,超精细,8k分辨率,战斗服的金属质感和她肌肤的光滑感以清晰度捕捉,增添了力量感和青春魅力




上图-Flux.1 Krea Dev / 下图-Midjourney V7


A graceful egret standing in a red and blue geometric space, full body shot, photorealistic, the bird’s elegant posture contrasted against the bold geometric shapes of the space, the red and blue tones of the background forming a dynamic and visually striking composition, soft natural light casting delicate shadows on the egret"s feathers, the smooth, pristine white of the bird standing out against the vibrant backdrop, hyperdetailed, 8k resolution, the textures of the egret’s feathers and the sharp, angular edges of the geometric space captured with clarity, creating a harmonious blend of nature and abstract art


一只优雅的白鹭站立在红蓝几何空间中,全身镜头,照片写实风格,鸟类优雅的姿态与空间大胆的几何形状形成对比,背景的红色和蓝色色调构成动态且视觉冲击力强的构图,柔和的自然光在白鹭羽毛上投下精致的阴影,鸟类光滑纯净的白色在鲜艳背景衬托下格外突出,超精细,8k分辨率,白鹭羽毛的质感和几何空间锐利的棱角边缘以清晰度捕捉,创造出自然与抽象艺术的和谐融合




上图-Flux.1 Krea Dev / 下图-Midjourney V7



A cute guinea pig piloting a small airplane, sitting confidently in the cockpit with its tiny paws gripping the controls, full scene shot, digital painting, the guinea pig wearing tiny aviator goggles and a leather cap, the cockpit detailed with buttons, levers, and dials, the airplane soaring through a clear blue sky with fluffy white clouds in the background, soft sunlight casting a warm glow on the scene, the texture of the guinea pig’s fur and the smooth metal of the airplane captured with playful charm, hyperdetailed, 8k resolution, the whimsical contrast between the small guinea pig and the large, mechanical airplane creating a fun and imaginative atmosphere

(太长了省略翻译了)


2.场景


这一轮我们主要测试图片创意和氛围感。




上图-Flux.1 Krea Dev / 下图-Midjourney V7


First-person view weaving between massive floating islands from the back of a dragon, waterfalls cascading into cloud seas, dragon scales reflecting sunlight and mist, rich fantasy colors, ultra-realistic fantasy style, high resolution, wide-angle lens


第一人称视角骑在巨龙背上穿梭于巨大的浮空岛屿之间,瀑布倾泻入云海,龙鳞反射着阳光和薄雾,丰富的奇幻色彩,超写实奇幻风格,高分辨率,广角镜头




上图-Flux.1 Krea Dev / 下图-Midjourney V7


An impressionist painting of ocean waves, the colors vibrant and swirling in a dynamic mix of blues, greens, and soft oranges, brushstrokes blending together to form the motion of the water, soft, blurred light and shadows playing across the surface of the waves, creating a dreamy and fluid atmosphere, the textures of the painting soft yet expressive, with the shimmering of light on the water"s surface gently implied rather than sharply defined, the overall effect capturing the fluidity and beauty of the ocean in a highly stylized manner, hyperdetailed, 8k resolution, the color palette rich and vivid, evoking the sensation of movement and the fleeting nature of the waves


一幅海浪的印象派绘画,色彩鲜艳生动,蓝色、绿色和柔和橙色动态交融旋转,笔触融合在一起形成水的运动,柔和模糊的光影在波浪表面流转,营造出梦幻而流动的氛围,绘画的质感柔和却富有表现力,水面上光线的闪烁被温柔地暗示而非锐利地描绘,整体效果以高度风格化的方式捕捉了海洋的流动性和美感,超精细,8k分辨率,色彩调板丰富鲜明,唤起运动的感觉和波浪转瞬即逝的本质




上图-Flux.1 Krea Dev / 下图-Midjourney V7


An underwater church ruin, ancient stone columns covered in swaying seaweed, full scene shot, photorealistic, beams of sunlight piercing through the water, illuminating schools of fish swimming around the ruins, the texture of the stone weathered and worn by time, soft waves of ocean current gently moving the seaweed, creating a mystical and serene atmosphere, dynamic shadows and light interplay on the sandy seabed, hyperdetailed, 8k resolution, the translucent water adding a dreamlike quality to the ancient structures, creating a harmonious blend of nature and history


一座水下教堂废墟,古老的石柱被摇摆的海草覆盖,全景镜头,照片级写实,阳光束穿透水面,照亮游弋在废墟周围的鱼群,石头的质感因时间而风化磨损,柔和的洋流轻柔地摆动着海草,营造出神秘而宁静的氛围,动态的阴影和光线在沙质海床上交相辉映,超精细,8k分辨率,半透明的海水为古老建筑增添了梦幻般的质感,创造出自然与历史的和谐融合




上图-Flux.1 Krea Dev / 下图-Midjourney V7


An orange glossy cartoon star-shaped sculpture standing atop a rounded green grassy hill, sunlight from the top-left, smooth reflective surface mirroring the blue sky, background of soft gradient light blue sky, low-angle close-up composition, realistic 3D render, high saturation colors, 35mm lens, f/4, high resolution, fresh and bright mood.


一个橙色光泽的卡通星形雕塑矗立在圆形绿色草丘顶部,阳光从左上方照射,光滑的反射表面映照着蓝天,背景是柔和渐变的浅蓝色天空,低角度特写构图,逼真的3D渲染,高饱和度色彩,35mm镜头,f/4光圈,高分辨率,清新明亮的氛围




上图-Flux.1 Krea Dev / 下图-Midjourney V7


A modern building with a white curved roof, its interior bathed in warm yellow lighting, full scene shot, photorealistic, the exterior features a gradient dusk sky, transitioning from deep oranges to soft purples, the water in front of the building reflecting its sharp, clean silhouette, the reflection perfectly mirroring the structure with subtle ripples, low-angle frontal composition, shot with a 35mm lens at f/5.6, high dynamic range rendering capturing the contrast between the building"s warm interior and the cool twilight, the serene and elegant atmosphere enhanced by the tranquil water surface and soft ambient light, hyperdetailed, 8k resolution, the textures of the building"s surface and the smoothness of the water clearly defined, the surrounding environment calm and inviting


一座拥有白色弧形屋顶的现代建筑,内部沐浴在温暖的黄色灯光中,全景镜头,照片级写实,外观呈现渐变的黄昏天空,从深橙色过渡到柔和的紫色,建筑前方的水面倒映着其锐利清晰的轮廓,倒影完美地映照着建筑结构并带有微妙的涟漪,低角度正面构图,使用35mm镜头f/5.6光圈拍摄,高动态范围渲染捕捉了建筑温暖内部与清凉暮色之间的对比,宁静优雅的氛围因平静的水面和柔和的环境光而得到增强,超精细,8k分辨率,建筑表面的质感和水面的光滑度清晰可见,周围环境宁静怡人




上图-Flux.1 Krea Dev / 下图-Midjourney V7


A pop art-inspired spring city park, vibrant colors with bold contrasts, full scene shot, digital painting, the park filled with stylized flowers, oversized trees, and geometric benches in bright, contrasting hues like electric pinks, yellows, and blues, the skyline in the background rendered in abstract, simplified shapes, soft clouds in the sky, playful details like oversized art sculptures and whimsical patterns in the grass, soft shadows and sharp light accents creating a dynamic atmosphere, mid-day sun casting vibrant light across the scene, hyperdetailed, 8k resolution, the combination of organic and geometric elements creates an energetic yet harmonious feel


一个波普艺术风格的春季城市公园,色彩鲜艳对比强烈,全景镜头,数字绘画,公园里充满了风格化的花朵、超大的树木和几何形长椅,采用明亮对比的色调如电光粉、黄色和蓝色,背景天际线以抽象简化的形状呈现,天空中飘着柔软的云朵,趣味细节如超大艺术雕塑和草地上的奇幻图案,柔和的阴影和锐利的光线点缀营造出动感氛围,正午阳光在场景中投射出鲜艳的光线,超精细,8k分辨率,有机与几何元素的结合创造出充满活力而又和谐的感觉




上图-Flux.1 Krea Dev / 下图-Midjourney V7


A steampunk-inspired industrial canyon, full of massive gears and towering smokestacks, full scene shot, concept art, the canyon is filled with rusted metal structures and pipes, the mechanical bridge in the background spanning across the canyon with intricate gears turning, steam billowing from the smokestacks, warm orange and copper tones dominate the scene, creating a vintage and gritty atmosphere, the sky is cloudy with hints of brown and gray, giving the whole scene a slightly dystopian, retro-futuristic feel, hyperdetailed, 8k resolution, the combination of industrial elements with artistic mechanical designs creating a blend of vintage and futuristic aesthetics


一个蒸汽朋克风格的工业峡谷,充满巨大齿轮和高耸烟囱,全景镜头,概念艺术,峡谷里布满生锈的金属结构和管道,背景中的机械桥梁横跨峡谷,复杂的齿轮在转动,蒸汽从烟囱中滚滚冒出,温暖的橙色和铜色调主导着整个场景,营造出复古粗犷的氛围,天空多云带有棕色和灰色色调,为整个场景增添了略显反乌托邦的复古未来主义感觉,超精细,8k分辨率,工业元素与艺术机械设计的结合创造出复古与未来主义美学的融合


3.特定风格


这一轮我们来看看更多不同风格属性要求下输出图片的表现。




上图-Flux.1 Krea Dev / 下图-Midjourney V7


A watercolor painting of Santorini, capturing the iconic white-washed buildings with blue domes perched on the cliffside overlooking the sparkling Aegean Sea, full scene shot, watercolor style, soft brushstrokes blending vibrant blues and whites with the surrounding natural landscape, the sea shimmering in delicate light, the architecture detailed with subtle shadows and textures, the sky a soft gradient from golden to light blue as the sun sets, gentle washes of color creating a dreamy and serene atmosphere, hyperdetailed, 8k resolution, the fluidity of the watercolor medium enhancing the peaceful, timeless beauty of the scene


一幅圣托里尼水彩画,捕捉标志性的白墙蓝顶建筑群坐落在悬崖边,俯瞰波光粼粼的爱琴海,全景镜头,水彩风格,柔和的笔触将鲜艳的蓝色和白色与周围自然景观融合,海水在精致光线下闪闪发光,建筑以微妙的阴影和质感呈现细节,天空呈现从金色到浅蓝色的柔和渐变,夕阳西下,温和的色彩晕染营造出梦幻宁静的氛围,超精细,8k分辨率,水彩媒介的流动性增强了场景的宁静、永恒之美




上图-Flux.1 Krea Dev / 下图-Midjourney V7


An oil painting of a noblewoman and her servant, set in an opulent 18th-century palace, the noblewoman dressed in an elaborate, richly embroidered gown with jewels adorning her neck and wrists, standing regally, while the servant, in simple yet neat attire, stands respectfully by her side, full scene shot, oil painting style, soft, warm lighting casting gentle shadows across their faces and clothing, the background filled with luxurious tapestries, golden accents, and velvet drapes, the textures of the fabric and skin rendered with meticulous detail, a rich and deep color palette of reds, golds, and soft creams, capturing the grandeur and elegance of the scene, hyperdetailed, 8k resolution, the brushstrokes adding depth and life to the aristocratic atmosphere


一幅描绘贵族女士和她的仆人的油画,背景设定在奢华的18世纪宫殿中,贵族女士身穿华丽的刺绣长袍,颈部和手腕佩戴珠宝,姿态高贵地站立,而仆人身着朴素却整洁的服装,恭敬地站在她身旁,全景镜头,油画风格,柔和温暖的光线在他们的面部和衣物上投下轻柔的阴影,背景充满奢华的挂毯、金色装饰和天鹅绒帷幔,织物和肌肤的质感以细致入微的细节呈现,丰富深邃的色彩调色板包含红色、金色和柔和的奶油色,捕捉场景的宏伟和优雅,超精细,8k分辨率,笔触为贵族氛围增添了深度和生命力




上图-Flux.1 Krea Dev / 下图-Midjourney V7


A cyberpunk-style flying vehicle soaring through a futuristic city, with a breathtaking and unconventional composition, full scene shot, concept art, the flying vehicle sleek and angular, glowing with neon lights in vibrant colors, against a sprawling cityscape with towering skyscrapers and dynamic neon signs, the sky filled with digital clouds and floating billboards, the background teeming with flying drones and holograms, dynamic motion blur effects capturing the speed and power of the vehicle, unconventional perspective, viewed from a low angle, emphasizing the height and futuristic nature of the city, high contrast lighting with glowing neon and dark shadows, hyperdetailed, 8k resolution, the entire scene exuding a sense of awe and advanced technology


一辆赛博朋克风格的飞行载具穿越未来城市,构图令人叹为观止且别具一格,全景镜头,概念艺术,飞行载具线条流畅棱角分明,散发着鲜艳色彩的霓虹光芒,背景是绵延的城市景观,高耸的摩天大楼和动感的霓虹招牌,天空中充满数字云朵和漂浮的广告牌,背景密布着飞行无人机和全息影像,动态运动模糊效果捕捉载具的速度和力量,非常规视角,从低角度观看,强调城市的高度和未来主义特质,高对比度照明搭配发光霓虹和深邃阴影,超精细,8k分辨率,整个场景散发出敬畏感和先进科技氛围




上图-Flux.1 Krea Dev / 下图-Midjourney V7


A warm and soothing flat illustration featuring a young girl with a gentle smile, her single eye peeking out from beneath a soft, oversized hat, with her reflection mirrored on a tranquil water surface, full-body composition, the color palette consisting of calming blocks of green, pink, and soft black, creating a peaceful, harmonious contrast, the lighting soft and inviting, casting delicate shadows that add depth, the girl"s expression filled with wonder and innocence, evoking a sense of warmth and comfort, the overall style is simple yet emotionally engaging, minimalistic with a touch of nostalgia, hyperdetailed, 8k resolution


一幅温暖舒缓的扁平插画,描绘一个面带温柔微笑的少女,她的单眼从柔软的超大帽子下探出,倒影映在宁静的水面上,全身构图,色彩调色板由舒缓的绿色、粉色和柔和黑色色块组成,营造出宁静和谐的对比,光线柔和温馨,投下精致的阴影增添深度,少女的表情充满好奇和纯真,唤起温暖和舒适感,整体风格简洁却富有情感感染力,极简主义中带有一丝怀旧气息,超精细,8k分辨率




上图-Flux.1 Krea Dev / 下图-Midjourney V7


A flat illustration with simplified shapes, featuring a single eye under a hat, with its reflection on a calm water surface, full scene shot, flat design style, the color palette consists of bold contrasts between green, pink, and black in solid color blocks, the reflection in the water echoing the geometric simplicity of the image, the eye glowing subtly, enhancing the mysterious and metaphorical feel of the composition, the mood is cold and enigmatic, with soft yet striking shadows adding depth, hyperdetailed, 8k resolution, the use of sharp lines and minimalistic elements contributing to a surreal, thought-provoking atmosphere


一幅采用简化形状的扁平插画,描绘帽子下的单眼,及其在平静水面上的倒影,全景镜头,扁平设计风格,色彩调色板由绿色、粉色和黑色纯色块之间的强烈对比组成,水中倒影呼应了画面的几何简洁性,眼睛微妙发光,增强了构图神秘和隐喻的感觉,氛围冷峻而神秘,柔和却引人注目的阴影增添深度,超精细,8k分辨率,锐利线条和极简主义元素的运用营造出超现实、发人深省的氛围




上图-Flux.1 Krea Dev / 下图-Midjourney V7


A 3D cartoon character of a father repairing a car in a garage, full body shot, stylized design, the father wearing overalls, with a tool belt full of wrenches and screwdrivers, kneeling down to fix the car’s engine, his facial expression focused yet kind, his body slightly bent as he works, the garage filled with mechanical tools, oil cans, and spare parts, warm lighting from overhead garage lights casting soft shadows, the car’s shiny surface reflecting the surroundings, vibrant colors of his clothing contrasting with the industrial tones of the garage, hyperdetailed, 8k resolution, the overall atmosphere is warm, hardworking, and full of love for the task at hand


一个3D卡通风格的父亲在车库里修车的角色,全身镜头,风格化设计,父亲穿着工装裤,腰间系着装满扳手和螺丝刀的工具带,跪下修理汽车引擎,面部表情专注而和善,身体微微弯曲工作着,车库里摆满了机械工具、油罐和备件,头顶车库灯光投射出温暖的光线和柔和的阴影,汽车光亮的表面反射着周围环境,他衣服的鲜艳色彩与车库的工业色调形成对比,超精细,8K分辨率,整体氛围温暖、勤劳,充满了对手头工作的热爱




左图-Flux.1 Krea Dev / 右图-Midjourney V7


Vintage tarot card illustration, black-and-white woodcut texture, a pair of lovers holding hands in a garden, an angel with spread wings behind them, roses in full bloom, sunrise in the distance, aged yellowed paper, labeled "LOVERS" at the bottom


复古塔罗牌插画,黑白木刻纹理,一对恋人在花园中手牵手,身后有一位展开翅膀的天使,盛开的玫瑰花,远处的日出,泛黄的陈旧纸张,底部标有"恋人"字样


4.其他




上图-Flux.1 Krea Dev / 下图-Midjourney V7


Hyper-realistic still life painting, icy blue crystal-like flowers intertwined with translucent orange candy vines, dark gradient background, fine material texture, lifelike gloss


超写实静物画,冰蓝色水晶般的花朵与半透明橙色糖果藤蔓交缠,深色渐变背景,精细的材质纹理,逼真的光泽




上图-Flux.1 Krea Dev / 下图-Midjourney V7


A flat illustration with simplified shapes, featuring a single eye under a hat, with its reflection on a calm water surface, full scene shot, flat design style, the color palette consists of bold contrasts between green, pink, and black in solid color blocks, the reflection in the water echoing the geometric simplicity of the image, the eye glowing subtly, enhancing the mysterious and metaphorical feel of the composition, the mood is cold and enigmatic, with soft yet striking shadows adding depth, hyperdetailed, 8k resolution, the use of sharp lines and minimalistic elements contributing to a surreal atmosphere


一幅采用简化形状的扁平插画,描绘帽子下的单眼,及其在平静水面上的倒影,全景镜头,扁平设计风格,色彩调色板由绿色、粉色和黑色纯色块之间的强烈对比组成,水中倒影呼应了画面的几何简洁性,眼睛微妙发光,增强了构图神秘和隐喻的感觉,氛围冷峻而神秘,柔和却引人注目的阴影增添深度,超精细,8k分辨率,锐利线条和极简主义元素的运用营造出超现实的氛围




上图-Flux.1 Krea Dev / 下图-Midjourney V7


Transparent purple candy-colored retro handheld game console, heart and star-shaped buttons, pastel pink background, soft light photography, sharp detail


透明紫色糖果色复古掌上游戏机,心形和星形按钮,柔和粉色背景,柔光摄影,锐利细节




上图-Flux.1 Krea Dev / 下图-Midjourney V7


An extreme close-up of an eye, the iris sharp and detailed, with a subtle reflection of a smartphone screen visible in the pupil, the reflection showing slight pixelation as the screen displays bright, vibrant colors, photorealistic, soft natural light highlighting the fine details of the eyelashes, eyelid, and skin around the eye, the sharp contrast between the smooth eye surface and the pixelated screen creating a striking visual effect, hyperdetailed, 8k resolution, the tiny reflections in the eye captured with extreme clarity, adding a sense of modern technology and intimate observation


眼部的极端特写,虹膜锐利且细节丰富,瞳孔中可见智能手机屏幕的微妙倒影,倒影显示出轻微的像素化效果,屏幕呈现明亮鲜艳的色彩,照片写实风格,柔和的自然光突出睫毛、眼睑和眼部周围肌肤的精细细节,光滑眼部表面与像素化屏幕之间的鲜明对比创造出引人注目的视觉效果,超精细,8k分辨率,眼中的微小倒影以极致清晰度捕捉,增添了现代科技感和亲密观察的氛围


小结


总体来说我个人感觉 Midjourney V7 还是细节更丰富质感更好许多,但是 Flux.1 Krea Dev 它开源啊。


Flux.1 Krea Dev 作为日常的图片生成和使用基本不成问题,但是如果要求更高质感的话还是优先 Midjourney。


之前有看到说它在视觉质量和风格上接近闭源模型如 FLUX.1 Pro ,这点上我是赞同的,上面可以看到输出图片在光影处理和纹理质感的渲染上都已经很不错了,肢体等方面虽然偶尔还会出错但是多 roll 两次还是能很快得到正常效果的输出图的。


在美学呈现上表现出色,人物塑造具有很强的真实感,并且它在扁平矢量插画风格上也有很不错的表现,3D质感渲染效果也还不错。对于一款免费开源的模型而言,这样的综合表现力已经非常难得了。


另外,它对提示词的遵循很好,复杂的提示词要求也大部分遵循了指令,并且输出了质量过关的图片,支持FP8量化,便于 LoRA 微调和自定义训练。在相对低端一些的硬件上同样可以运行。


好啦今天的分享就到这里!Qwen-Image 我下载到本地了,不过下期几时能写出来还不确定哈哈哈,期待大家的三连猛猛催更,我会继续努力给大家带来认真的测评和内容的!


文章来自于微信公众号“阿真Irene”,作者是“宝藏同学阿真”。


作者头像

AI前线

专注人工智能前沿技术报道,深入解析AI发展趋势与应用场景

246篇文章 1.2M阅读 56.3k粉丝

评论 (128)

用户头像

AI爱好者

2小时前

这个更新太令人期待了!视频分析功能将极大扩展AI的应用场景,特别是在教育和内容创作领域。

用户头像

开发者小明

昨天

有没有人测试过新的API响应速度?我们正在开发一个实时视频分析应用,非常关注性能表现。

作者头像

AI前线 作者

12小时前

我们测试的平均响应时间在300ms左右,比上一代快了很多,适合实时应用场景。

用户头像

科技观察家

3天前

GPT-4的视频处理能力已经接近专业级水平,这可能会对内容审核、视频编辑等行业产生颠覆性影响。期待看到更多创新应用!