GPT-6 Astra 3D 动画工作室构建

一位开发者仅用大约一周的提示工程时间，就搭建了自己的 3D 动画工作室。 四个实时视口、残影轨迹、79 关节骨骼绑定、85 个动作片段库。全部由 GPT-6 Astra 制作完成。 奇怪的是：他甚至还不知道这些素材将用于哪款游戏。 观察每个角色身后拖曳的淡化副本。每一个残影都是之前的帧，因此整个动作同时呈现在屏幕上。0:12 处的有角恶魔最能体现这一点。 该流程将两个免费来源拼接在一起 → 使用 Mixamo 获取标准动作，使用 Kimodo（NVIDIA 的开源动作模型）处理中间过渡。 随后他分享了这一技巧。 截取一个待机姿势的单帧。将其设为第一帧和最后一帧。在中间放置一个人物躺在地上的姿势，并在 2 秒内输入提示词“person rolling on the ground to the right”。 结果是一个《黑暗之魂》风格的翻滚动作，起始和结束姿势完全相同。只需在提示词中改变方向，即可得到另一侧的翻滚。 成本是多少？大致相当于 Lite 版本模型的一周使用量。本月早些时候，他用同样的 AI 在几天内完成了一款完整的游戏。 这款游戏尚不存在。但为其打造的工作室已经就绪。 需要明确的是，你无法下载此工具。它是围绕他自己的工作流程构建的，他表示 AI 并不是难点。让 UI 感觉合乎逻辑才是。 独立开发者选择自建工具而非购买：这是新常态，还是阻碍游戏发布的无底洞？

One dev built his own 3D animation studio in about a week of prompting. Four live viewports, ghost trails, 79-joint rigs, an 85-clip motion library. All made with GPT-6 Astra. The strange part: he has no idea what game it's for yet. Watch the faded copies trailing each character. Every ghost is an earlier frame, so the whole move sits on screen at once. The horned demon at 0:12 shows it best. The pipeline is two free sources glued together → Mixamo for stock motion, Kimodo (NVIDIA's open motion model) for everything in between. Then he shared the trick. Grab one frame of an idle pose. Set it as both the first and the last frame. Put a pose of someone on the floor in the middle and prompt "person rolling on the ground to the right" over 2 seconds. Out comes a Dark Souls style dodge roll that starts and ends in the exact same stance. Flip the direction in the prompt and you have the other side. The cost? Roughly one week of usage on the Lite version of the model. Earlier this month he used the same AI to finish a whole game in a few days. The game doesn't exist yet. The studio for it already does. To be clear, you can't download this. It's built around his own workflow, and he says the AI wasn't the hard part. Making the UI feel logical was. Solo devs building their own tools instead of buying them: the new normal, or a rabbit hole that stops games from shipping?