Flux 2开源 320亿参数量对标Nano Banana Pro Flux 2 Lora训练 附工作流

FLUX.2 Pro

闭源顶级版

★★★★★

1

商业 API

品牌广告、影视预览、出版级

FLUX.2 Flex

半开源高配

★★★★☆

2

API + 权重（非商业）

设计师、快速迭代

FLUX.2 Dev

开源主力

★★★★

3

Apache 2.0

本地部署、二次开发

FLUX.2 Klein

轻量骨干

★★★★★

4

Apache 2.0

研究、边缘设备、超高并发

"A close-up portrait of a seasoned female journalist in her late 50s, with short silver hair and sharp, intelligent eyes behind rimless glasses, wearing a crisp white blouse with rolled sleeves, subtle laugh lines around her mouth conveying wisdom and warmth, soft window light from the left casting gentle shadows, professional headshot style, high resolution, 8K."

聚焦面部特征 + 光影描述，避免关键词堆砌。

Anakin AI 指南：生成情感深度肖像，解剖准确率 98%。

"Dark-skinned woman, mid-20s, close-up portrait, with a thoughtful and contemplative expression. Her eyes are covered by charred, torn pieces of aged parchment with text visible. Gold flakes and dust are incorporated into the texture, giving an intricate and layered look."

纹理细节 + 叙事元素。

PromptHero 热门：艺术写实融合，获 500+ 点赞。

"超写実的なポートレート、若い女性、自然な表情、柔らかな視線、肩までの構図、映画のようなライティング、浅い被写界深度、滑らかな肌の質感、プロフェッショナル写真、ソフトな自然光、ミニマルな背景、落ち着いた色調、高解像度、スタジオ撮影"（日语提示）。

多语言支持，直接用母语描述。

X 用户 @azed_ai：跨文化肖像，Flux 2 无需翻译层。

"POV of an astronaut holding a small red rock that has a portal showing the Statue of Liberty, alien landscape of a red planet in the background, surreal, ultra-realistic, photorealistic."

第一人称视角 + 超现实对比。

AgeofLLMs 示例：梦幻门户效果，Reddit 热议。

"Wide angle shot of eyes off to one side of frame, lucid dream-like 3D model of owl, game asset, blender, looking off in distance ::8 style | glowing ::8 background | forest, vivid neon wonderland, particles, blue, green, orange ::7 parameters | rule of thirds, golden ratio, asymmetric composition, hyper-maximalist, octane render, photorealism, cinematic realism, unreal engine, 8K ::7 --ar 16:9 --s 1000."

权重 (::8) + 构图规则（黄金分割）。

3DAI Studio 博客：霓虹森林猫头鹰，游戏资产生成。

"3D model of a green war balloon, clash of clans, fantasy game, front view, game asset, detailed, war ready, photorealistic, in a war environment, spring, Disney style, Pixar style."

风格混合（Disney + Pixar）。

3DAI Studio：战争气球，迭代编辑友好。

"Cinematic scene, over the shoulder close-up portrait of a soldier on the front line, intense and mesmerizing look, as explosions are seen in the background, film grain, depth of field."

肩后视角 + 动态元素（爆炸）。

AgeofLLMs：战争电影海报，粒噪真实。

"Hyperrealistic photo of an appetizing piece of perfect polished cake with caramel filling, a beautiful perfect piece of cake lies on a tiny, simple caramel or beige dessert plate, behind this piece of cake in the background we see the same cake but is round, creamy, simple dessert plate, beige background with no additives, perfect cakes without any crumbs as in the advertisement, the photo has a beautiful and clean composition, all shown on white background, beautiful composition, award-winning quality photo, professional photo, highly detailed, bokeh behind, professional composition, daylight, warm atmosphere, cinematic look, sharp photo, high contrast, detailed photo."

背景虚化（bokeh） + 广告级细节。

Medium 测试：食物摄影，Flux 胜 Midjourney。

"A couple home portrait, girl with long beautiful black hair, boy with dry sophisticated short brown hair, both wearing loose and comfortable blue and white striped shorts, girl tank top and boy short sleeves, both sitting comfortably on the sofa, close-up figures, setting in cozy living room, soft interior light, artistic photography, portrait photography, high quality, 8K."

服装/姿势细节 + 室内光。

Medium 9 提示测试：家庭肖像，一致性高。

{ "subject": "Young Asian woman with flawlessly sleek jet-black hair framing her face, wearing a shimmering black organza dress and a delicate silver bracelet with embossed lettering, soft rose lips", "action": "Looking with a confident, direct gaze", "style": "High-end fashion photography, dramatic studio portrait, Paolo Roversi × Nadav Kander aesthetic, high tonal contrast", "context": { "environment": "Professional studio", "background": "Dark gray backdrop with subtle silver reflection", "mood": "Timeless sophistication, quiet, elegant stillness", "lighting": "Soft directional light, softbox key light, velvety shadows, gentle rim glow, controlled falloff" }, "color_palette": ["Deep Charcoal", "Muted Silver", "Jet Black", "Warm Skin Highlights", "Soft Rose"], "composition": "Portrait framing, emphasis on texture and lighting", "camera": { "angle": "Eye level", "shot": "Medium shot", "lens": "85mm portrait lens", "aperture": "f/2.8", "ISO": "100", "focus": "Sharp focus on eyes" } }

模块化（主体 + 相机参数）。

X @azed_ai 线程：时尚肖像，颜色匹配 100%。

{ "subject": "Male fashion model wearing an avant-garde translucent raincoat with a gradient from Neon Green #02eb3c to Lemon Yellow #edfa3c, Deep Royal Purple #2a0052 cargo trousers, and Hot Pink #ff0088 sneakers", "action": "Walking confidently down a runway", "style": "High-end streetwear photography, vibrant color blocking, sharp focus", "context": { "environment": "Minimalist fashion show", "background": "Clean concrete wall", "mood": "Energetic, futuristic, bold", "lighting": "Bright studio lighting, soft shadows" }, "color_palette": ["#02eb3c (Neon Green)", "#edfa3c (Lemon Yellow)", "#ff0088 (Hot Pink)", "#2a0052 (Deep Purple)"], "composition": "Full body shot, low angle to emphasize the outfit", "camera": { "angle": "Low angle", "shot": "Full body", "lens": "35mm", "aperture": "f/8", "ISO": "100", "focus": "Sharp focus on the clothing textures" } }

HEX 码 + 渐变控制。

X @azed_ai：街头时尚，HEX 精度高。

"Generate an infographic about the perception of 'AI Slop.' A gritty, dark-mode editorial infographic displaying 'The Anatomy of Digital Slop'. The layout flows vertically from top to bottom. In the top header section, a deconstructed flow chart shows 'The Feedback Loop' with arrows forming a recycling symbol made of distorted, pixelated JPEGs. In the middle section, a comparative split-screen presents 'Soul vs. Syntax' in harsh neon green and grey, contrasting a messy, textured human brushstroke against a smooth, overly polished, plastic-looking AI surface. At the bottom, a glitch-art bar graph illustrates 'The Uncanny Valley Effect' using melting, identical faces that fade into digital noise. The design features a 'Cyberpunk Dystopia' palette of deep charcoal, toxic neon green, and warning red, using bold, distorted glitch-style fonts for headings and mono-spaced code font for body text. Matte black background with static grain texture and harsh, cold screen lighting. The overall composition emphasizes the feeling of artificiality and mass production through repetition and visual dissonance, creating a cynical, rebellious aesthetic."（可转 JSON）。

数据可视化 + 赛博朋克风格。

X @azed_ai：信息图单次生成，布局精确。

"A [Style] sticker of a cute Red Panda, featuring soft gradient shading, expressive eyes, and holding a bamboo shoot with a vibrant organic color palette of Rust Red #B93B18, Cream White #FFFDD0, and Bamboo Green #789640. The design should have delicate ink outlines with a rough paper grain texture, artistic paint bleeds, and a thick white die-cut border."（[Style] 替换为 watercolor/ink 等）。

变量风格 + 纹理（纸张颗粒）。

X @azed_ai：贴纸套装，艺术表达强。

"Digital illustration of a crime movie poster 'The Codfather', showing a stern-looking fish in a suit, sitting behind a desk made of coral, with other fish characters lurking in the shadowy underwater background, movie title in a gritty font, intense color palette."

幽默叙事 + 海报构图。

AgeofLLMs：水下黑帮，Schnell 变体优化。

"How would a 5 year old kid draw a portrait of Steve Buscemi? Childlike crayon drawing, simplistic lines, exaggerated features, vibrant primary colors, innocent and playful style."

抽象儿童画模拟。

X @techhalla：抽象肖像，Flux 胜 Nano Banana。

Rectified Flow Transformer

空间与组成建模，生成流动路径

从Flux.1的简单Flow Matching升级为Rectified版本，提升物理一致性20%

~8B（骨干）

Mistral-3 VLM

提示解析与世界知识注入

新增24B VLM融合，提示遵循Elo分数提升15%

24B

VAE Encoder/Decoder

潜在空间压缩与解码

从头重训，解决三难困境，支持4MP原生输出

~0.1B