重點導讀
Archon 是 Selendis AI 的影片基礎模型,可對原始素材進行零樣本剪輯,並持續學習個別創作者的剪輯偏好。
- 偏好訊號:創作者在 Selendis 多人編輯器中的決策、過往影片作品,以及引用的參考影片。
- 學習方式:Archon 將這些訊號視為依創作者而異的偏好先驗,並納入接受或拒絕等編輯決策。
AI 伴隨問答
針對這篇報導有疑問?可直接點選快捷問題或向 AI 追問背景脈絡與技術細節:
Rohan Paul@rohanpaul_ai
·19 天前
Creative tools are starting to learn taste from behaviour instead of from prompts.
Archon, the video foundation model from @SelendisAI, takes raw footage and returns a finished edit zero-shot, then keeps learning how a specific creator likes to cut.
The taste signal comes from three places:
- decisions made inside Selendis’s multiplayer editor
- the creator’s past video catalogue
- reference videos the creator cites
i.e. Archon treats taste as a learnable per-user prior, conditioned on a creator’s back catalogue, cited reference videos and accept/reject decisions inside the editor.
推文來源 · X (Twitter)在 X 參與討論
來源: Rohan Paul (x.com)
在 X 查看推文