微信扫码
与创始人交个朋友
我要投稿
SD3模型发布之前,全网对它充满了期待。
然而,正式发布后,我用了几个小时肝到半夜1点调通工作流后,发现生成的图片质量比较有限。
不过,SDXL和SD1.5的强大之处在于开源社区的众网友的努力,SDXL的原生模型也是表现平平。
所以用SD3的时候,应该和SDXL的原生模型对比,而不是和创作者们改造的大模型对比,更何况SD3放出来的是“中杯”,完整版模型尚未放出。
让AI绘画爱好者们饱受诟病的是,SD3模型内置了对nsfw的限制,结果导致一些特定场景无法画出正常的躯体。
比如这样:
只要画草坪上的人,必然会画崩。
作为开源的本地模型,不让涩涩是什么道理?!
发布之前,通过API,大家体验到了超越MJ的画质;发布之后,通过本地大模型,大家体验到了什么叫翻车。
几乎同时,基于SDXL的HelloWorld7.0发布了。
网友评论:是不是这才是我们等待的真·SD3?
让我们感受下这个最优秀的真人大模型之一的效果。
1、蓝色围巾
提示词:
cold-weather portrait of a young japanese woman with tousled hair clutching her head, wearing a blue scarf and gray blazer, large eyes and soft facial features, 18 years old, realism photography, cool color palette with a soft focus, indirectly inspired by the aesthetics of wong kar wai, masterpiece
2、玉兰花
提示词:
close-up portrait of a young chinese woman in traditional attire holding white magnolia flowers, clear skin, delicate features, long black hair, elegant embroidered white dress, gentle expression, 18-22 years old, inspired by zhang jingna, chinese ink painting style, floral elements, fine detail, masterpiece
3、当然了,不止能画中国女孩,欧美男生也可以
提示词:
half-body image of a young man with intense orange eyes and fresh facial cuts, wearing a black hooded cloak, short brown hair, 28 years old, pale skin, dark dramatic lighting emphasizing his muscles, gothic horror aesthetic, photographic realism, facial hair, detailed texture, inspired by caravaggio in chiaroscuro style, masterpiece,
4、汽车
提示词:
shot of a sleek, modern car inside a dark garage, headlights shining brightly, dramatic lighting, low-angle perspective, monochromatic tones, the car's front is the focal point, haze and mist in the surroundings, minimalist composition, futuristic design, high contrast between light and shadow, inspired by elon musk's tesla designs, best quality,
5、软科幻场景
提示词:
post-apocalyptic anthropomorphic kangaroo in advanced combat armor gear, middle-aged, anthropomorphic character, digital painting, dark palette, sci-fi theme, heavy mechanical elements with cybernetic enhancements, grayscale background, rugged texture, conceptual art inspired by moebius aesthetics, masterpiece
6、汉服
提示词:
elegant portrait of a young asian woman in traditional chinese attire, serene expression, delicate hair ornaments, soft blue robe with intricate embroidery, hands gently crossed, 18 years old, film grain effect, dreamy lighting, inspired by shen quan's portraits, masterpiece
7、动漫风格恋爱剧情
提示词:
intimate anime-style interaction between a young couple, the girl with long black hair whispers passionately to a flustered boy, blush on cheeks, white shirt, black t-shirt, stylish digital illustration with deep shadows and dramatic lighting, inspired by shoujo manga, 17-18 years old, japanese animation, masterpiece
8、少数民族
提示词:
young woman in ornate traditional costume, eyes closed, hands in prayer, complex golden headdress with ruby-like gems, pearl and gold chain face veil, blue and gold garments, wrist adorned with gold bracelets, serene expression, approximately 10 years old, east asian, high-resolution digital photograph with soft focus, masterpiece
9、大猫
提示词:
adorable tiger cub portrait, detailed close-up, resting head on front paws, golden-brown fur with black stripes, large expressive eyes, pink nose, fluffy ears, realistic photography, young tiger approximately 3 months old, natural habitat blurred in the background, inspired by wildlife photography, masterpiece
10、静物
提示词:
Masterpiece, RAW photo, ultra-high resolution, absurdres, (color pencil drawing:1.3), Imagine a cozy morning scene with a steaming coffee cup, a freshly baked croissant, and a vase of vibrant flowers on a rustic wooden table. Describe the atmosphere, the aroma, and how this setting makes you feel.volumetric light, Sharp Focus, 8K, 16K, UHD, hyperrealistic, masterpiece, high quality, wide depth of field, Nikon D850, extremely detailed, high angle shot,
模型下载:
https://pan.quark.cn/s/054cab5ff321
53AI,企业落地应用大模型首选服务商
产品:大模型应用平台+智能体定制开发+落地咨询服务
承诺:先做场景POC验证,看到效果再签署服务协议。零风险落地应用大模型,已交付160+中大型企业
2024-11-15
西湖大学&腾讯:一个多模态Web Agent的开源框架
2024-11-13
最复杂多智能体发布!百度推出“秒哒”和文心iRAG
2024-11-12
【RAG&多模态】多模态RAG-ColPali:使用视觉语言模型实现高效的文档检索
2024-11-11
开摆!谷歌AI视频上线!脚本、素材、剪片全稿定!
2024-11-11
文档OCR版式识别,兼顾速度与精度,YOLO当首选
2024-11-10
硬核升级!在Ollama中使用Llama3.2视觉模型
2024-11-08
dify案例分享-基于多模态模型的发票识别2-多种发票识别
2024-11-07
星辰 AI 大模型:中国电信的 AI 大模型集合平台,支持多模态任务和多语种处理
2024-05-30
2024-09-12
2024-06-17
2024-08-06
2024-08-30
2024-04-21
2024-06-26
2024-07-07
2024-06-14
2024-07-21
2024-09-26
2024-09-26
2024-09-01
2024-07-15
2024-07-14
2024-07-10
2024-07-02
2024-06-29