8 月 20 日消息，科技媒体 testingcatalog 昨日（8 月 19 日）发布博文，报道称 Meta 公司即将推出 Muse Video 模型，目前已邀请部分合作伙伴 Beta 测试，单次可以生成最长 10 秒视频。
根据该媒体测试生成的视频，Muse Video 在细节呈现、对场景和物体关系的理解，以及连续画面在时间维度上的稳定性方面，展现出较高潜力。
Meta 此前承认模型仍存在音频与画面同步不足的问题，对高速运动的物理准确性也有差距。现有样本为 10 秒片段，尚不足以验证模型在更长时长叙事、复杂角色持续一致性或高强度动作场景中的稳定表现。
附上相关测试生成的视频如下：
提示词“A woman with curly red hair and a green scarf walks toward camera down a crowded Tokyo street at night, stops, and looks up”（一位有着红色卷发、围着绿色围巾的女子在夜晚拥挤的东京街道上走向镜头，她停下脚步，抬头望向前方。）A shaky, low-quality iPhone video recording. A guy holding the camera selfie-style walks up to random people on a busy city street and asks each one, "How do you feel about being AI generated?" The video has authentic smartphone artifacts—slight motion blur, overexposed sky, wind noise, and the occasional finger over the lens."“一段抖动且画质很差的 iPhone 视频。一个男人拿着手机，像自拍一样，走到繁忙的城市街道上随机问路人：“你对被人工智能生成有什么感觉？”视频中存在典型的智能手机拍摄痕迹 —— 轻微的运动模糊、过曝的天空、风噪，以及偶尔出现的遮挡镜头的手指。”"The rocket at the rocket station is launching. Everything looks normal until the rocket explodes into confetti pieces and leaves nothing behind."“火箭发射站的火箭正在发射。一切看起来都很正常，直到火箭爆炸成碎片，什么也没留下。”"The man couldn't believe it when he finally became a kitten dad. Excitedly he shouted "I'm a dad now! A cat dad!". He's so happy that he starts tearing up. The kitten meows before the video ends."“当他终于当上猫爸时，他简直不敢相信自己的眼睛。他兴奋地大喊：‘我当爸爸啦！猫爸！’他高兴得热泪盈眶。视频结束前，小猫喵喵叫了一声。”"A glass of red wine tips over on a white marble counter, spilling toward the camera. Static shot" 一杯红酒倾倒在白色大理石台面上，酒液溅向镜头。固定镜头。
"A cat made out of Lego technic in the rain made out of fluffy plastic rain"“一只用乐高科技系列积木拼成的猫，在蓬松塑料雨中”"Orbiting shot around a chef plating a dish, 180-degree arc, shallow depth of field."“围绕正在摆盘的厨师进行 180 度弧线环绕拍摄，景深较浅。”"A neon sign reading ‘OPEN 24 HOURS’ flickers on above a diner door in the rain." 雨中，一家餐馆门上方的霓虹灯招牌闪烁着“24 小时营业”的字样。"A car’s headlights sweep across a bedroom wall at night, shadows shifting."“夜里，汽车前灯扫过卧室墙壁，影子不断移动。”Two swordsmen in a bamboo forest exchange three strikes, blades meeting mid-frame. Wide shot, overcast light.两名剑客在竹林中交锋三下，刀刃在画面中央相撞。远景镜头，阴天光线。"Two fighters in a dimly lit warehouse; one throws a punch, the other blocks and counters. Handheld camera, shallow depth of field."“两名拳击手在光线昏暗的仓库里对打；一人出拳，另一人格挡反击。手持摄像机，浅景深。”"Cyberpunk hacker robot working in front of many monitors"“赛博朋克黑客机器人正在多个显示器前工作”
AI 中文社
