#Sora2

群里看到的一个提示词,分享下。 这个汽车的三视图,牛了 生成的图直接喂给Sora2、Veo3.1等大模型 可试试直出一致性的视频效果 做视频的效率又提升了。 NanoBanana提示词: Prompt : A cinematic automotive photoshoot of me, keeping my real face unchanged. The scene is composed of three perspectives as a : 1. Top Panel: Interior close-up: I am seated inside the car, wearing a fitted black polo shirt, shown in the side mirror reflection. My face is serious and focused, my gaze directed forward with determination. The angle captures only side view part of my profile, framed cleanly within the mirror, emphasizing intensity and precision. 2.Middle Panel: I am standing confidently next to a sleek black Ford Mustang. My posture is relaxed but strong: both arms are crossed over my chest in a confident manner, while my left leg is straight and my right leg bent slightly at the knee, with the foot leaning casually against the car. My gaze is directed slightly off-camera, with a calm and assertive expression. I am wearing a fitted black polo shirt with subtle detailing, slim grey jeans with a clean cut, and brown leather boots. My outfit is minimal yet stylish, emphasizing a modern masculine vibe. Pose like a pro, same face as the uploaded photo. 3. Bottom Panel: Rear car shot: The camera captures the back of the Mustang, showcasing the "YOUR NAME" license plate and muscular lines of the car. The photo emphasizes the glossy texture of the vehicle and its aggressive, cinematic presence. The setting is an urban environment with modern architecture and concrete walls, giving a gritty, cinematic atmosphere. The lighting is natural but slightly diffused, highlighting both me and the polished surface of the car. The perspectives vary: - The mirror reflection close-up is shot tight with a portrait focal length (~85mm). - The exterior full-body shot is taken at eye-level with a slightly wide lens to capture both me and the car in full view. - The rear car angle uses a low perspective to emphasize power and presence. Style: Cinematic automotive editorial, urban setting, moody and stylish, professional fashion-meets- car photography, same face.
Sora2的服化道还是可以的😁 朱墙宫怨 prompt: video_attributes: total_duration: 15s frame_rate: "24fps" film_grain: "无颗粒,追求极致清晰和质感的数字电影感(Arri Alexa 风格)" tone: "华丽、压抑、庄重、暗流涌动、富有戏剧张力" color_palette: "深红(宫墙)、金色(刺绣)、深木色。低调奢华,高对比度,阴影部分丰富,整体偏暖色调。" audio: ambient: "寂静的宫殿,远处隐约的更夫打更声,烛火燃烧的轻微噼啪声。" music: "缓慢、沉重的弦乐(大提琴)和琵琶(Pipa)的旋律,营造紧张和悲凉感。" sequence: - shot_1: duration: "6s" composition: "中景(Medium Shot),对称构图,使用 35mm 变形镜头(Anamorphic)。" camera_motion: "极其缓慢的轨道后拉(Dolly out),从她的面部特写开始,逐渐拉开,展现环境。" lighting: "模拟烛光的暖色调光线,从侧面照亮她的脸,另一半脸陷入阴影。强烈的伦勃朗光。" subject: description: "一位面容精致的妃子(清代设定),妆容无可挑剔,柳叶眉,眼神复杂,似有悲伤和不甘。" wardrobe: "一件极其奢华的深紫色丝绸旗装,上面有精美的金色凤凰刺绣,佩戴着沉重的点翠和珍珠头饰(钿子)。" scene: location: "紫禁城(或类似)的华丽寝宫内部。" time_of_day: "深夜,大约子时。" environment: "背景是雕花的深色木质屏风,桌上摆着黄铜鹤式烛台和珠宝盒。空气中似乎有熏香的薄烟。" visual_details: action: "她静静地坐在梳妆台前,手中拿着一支金步摇,但目光却空洞地望向跳动的烛火。" props: "黄铜烛台(火焰跳动),红木梳妆台,金步摇。" transition_to_next: "硬切 (Hard Cut)" - shot_2: duration: "4s" composition: "特写(Close-up),焦点在她的眼睛和头饰上,85mm 镜头。" camera_motion: "固定机位(Static)。" lighting: "烛光在她的瞳孔中反射出微小的光点,点翠头饰在暗光下依然闪耀。" subject: description: "她的眼睛特写,睫毛微颤。" wardrobe: "点翠头饰的羽毛和宝石细节。" scene: location: "同上。" time_of_day: "深夜。" environment: "背景是模糊的屏风图案。" visual_details: action: "她缓缓眨眼,一滴眼泪从眼角滑落,划过精致的妆容。她没有擦拭。" props: "一滴清晰的眼泪。" transition_to_next: "L-cut (音乐延续)" - shot_3: duration: "5s" composition: "过肩镜头(Over-the-shoulder),从她身后拍摄,看向窗外。" camera_motion: "非常缓慢的推近(Slow push-in),增加压迫感。" lighting: "来自窗外的清冷月光(蓝色调)与室内的暖烛光(橙色调)形成鲜明对比。她的背影被月光勾勒出轮廓。" subject: description: "她的背影,头饰的轮廓。" wardrobe: "旗装背部的刺绣细节。" scene: location: "同上,但朝向窗户。" time_of_day: "深夜。" environment: "窗外是深蓝色的夜空和一轮明月,以及宫殿屋檐的剪影。" visual_details: action: "她站起来,背对镜头,凝视着月亮。宫殿的阴影笼罩着她。" props: "雕花的窗棂。" transition_to_next: "淡出到黑色 (Fade to black)"
赵纯想
1个月前
媒体对Agent Builder不兴奋,对Sora2很兴奋。原因是猴性太重,和普遍C端一样,只喜欢能刺激眼球的东西。 Agent Builder不是Coze,不是Dify。它不是工作流的编排和演绎。工作流压根儿、从来就不是Agent,因为它只有固定的流向、固定的产出物。而OpenAI的拖拉拽面板,不是让你规划工作流用的。而是对Agent装配的一种抽象。我花了三个月,探索ClaudeCode的逆向库,才用Go复刻完成的一种Agent的装配,现在所有开发者只需要动动手指就能得到。这种抽象带来的正是Agent核心封装技术的下放和普惠。 Think + ToolUse的排列组合,与固定工作流不同,它代表无穷的可能性。是LLM自身决定下一步该做什么。是真正的Agent,就像你手边的ClaudeCode 和 GeminiCLI。观察你常用的CodingCLI的工具调用链路,每一次都不是固定的。未来,结合你自身的业务设计一系列的工具,由LLM在思考后自身决定调用和调用顺序,就可以释放巨大的智能。而OpenAI,将这一切可视化了。 这还不是重点,重点是OpenAI还想吃下整个交互侧的前端实践。配合Chatkit的Widgets生成能力,我在20秒之内得到了对话流中的交互式组件。将相关组件添加到Agent的体系中,就能实现与用户的垂直场景客制化Agent。每个场景都有自身的专属UIUX,不再是简单的一次性工作后返回,而是将一切App都变身为Cursor的潜力。 图片中就是我自己在laper中设计了很久的对话式故事探讨UIUX交互的OpenAI实践,20秒,颠覆了2个月以来的复杂工作和设计。有句话说得好,"未来已经到来,只不过分布不均"。