跳到正文
DEV Community· AI-KSK·· 7 天前AI 評分50

AI-KSK 分享 SCAIL 2 長篇參考編輯的四套工作流

SCAIL 2 Long-Form Editing: Four Reference Workflows

AI 導讀

AI-KSK 分享 SCAIL 2 長篇參考編輯的四套工作流,涵蓋單人、雙人、單一生物及雙生物的長影片處理。工作流在每段邊界重新錨定參考,處理影片變長後參考漂移、身份弱化及身體比例偏移的問題。雙主體版本保留兩條獨立條件流,直到 sampler 才合併;作者亦提供工作流 JSON 和教學影片連結。

正文

AI-KSK

Four SCAIL 2 workflows focused on long-form reference editing — single and dual subjects, people and creatures.

The four workflows

# Workflow Subjects
1 Single-reference edit, long video 1 person
2 Single-creature long-video drive 1 creature
3 Dual-reference edit, long video 2 people
4 Dual-creature long-video drive 2 creatures

The long-video problem

Reference editing works well on short clips and gets harder as length grows: the reference drifts,
the identity softens, and body proportions wander. These four graphs handle it by re-anchoring on
the reference at segment boundaries rather than conditioning once at the start.

Dual-subject is the hard case

Two people means two identities that must not merge. The dual-reference graphs keep separate
conditioning streams and only merge them at the sampler, which is why they survive longer than a
single combined prompt.

Running Hub links

Files

Workflow JSONs and covers: https://huggingface.co/AI-KSK/ksk-video-261-comfyui-workflows

Video walkthrough (Chinese): https://youtu.be/_LbRVl9hwnI

Bilibili: https://www.bilibili.com/video/BV1w2Ei6pEsJ/

Notes

  • ComfyUI exports that reference models present in the cloud instance; not self-contained packs.
  • Tutorial material, not a product.
  • Independent AI-KSK release. Not affiliated with or endorsed by the model authors or RunningHub.

來源:DEV Community · dev.to