專案

建立專案
管理員切換登入身分，權限矩陣會即時套用權限矩陣中「核准腳本／片段／成片 — Editor：視設定」的那個設定
清空 localStorage，把 mock 資料還原成初始狀態
專案/

資金從比特幣輪動到生態代幣

腳本待審
周曉薇@meme_coin_diary目標 58 秒4 個片段成本 $0.05負責人 柯昱翔更新 2026/07/30 03:40
前往腳本編輯
攝取腳本音檔動畫合成
36%
人工閘門擋下推進卡在「腳本已核准」：腳本閘門未通過：1 項待修正（seg_0403：84 字，超出 4 字（目標 75–80 字））字數 75–80、禁用詞與雙版本逐字一致三項全過才可核准鎖版（spec 第 5 節）。這一關不會被快轉：改的是口播內容，只能由人決定。前往腳本編輯 →

Higgsfield 任務

半自動任務流程：系統負責把 Prompt、素材與檢查清單準備到可以直接照做， 人只做 Higgsfield 上真正需要人的那一步。

專案狀態腳本待審

這一版刻意不做瀏覽器機器人

Higgsfield 的生成品質需要人眼判斷，替它寫一支會壞掉的自動化腳本只是把風險藏起來。 我們把可以自動化的部分做到底：Prompt 自動帶入正確的主持人、背景、構圖與禁止事項， 素材與檔名一次備妥，上傳強制綁定 segment_id。人負責的只剩「看一眼、按生成、丟回來」， 而且每一步都留在流程裡可追蹤、可重做。

批次進度

需要動畫的段落

3

音檔就緒

0

已下載任務包

0

已上傳成果

0

待處理

3

段 1・待生成段 3・待生成段 4・待生成
段 1seg_0401主持人全畫面已驗證待生成spec 版本 A 第 7 節：每段狀態顯示待生成 / 已上傳 / 待審核 / 核准 / 退回重生
音檔尚未就緒：請先到「音檔生成」分頁完成本段 Cartesia 音檔，任務包才會有對嘴音軌。
host_reference.png
background.png

周曉薇

@meme_coin_diary
ctv_zh_tw_female_meme_04

音檔尚未就緒：請先到「音檔生成」分頁完成本段 Cartesia 音檔，任務包才會有對嘴音軌。

Higgsfield Prompt15.0sproj_04_seg_0401_v1.mp4

[HOST] A playful Taiwanese female creator in her mid-20s, high ponytail, oversized graphic tee, bouncing slightly in a neon-lit bedroom studio with LED strips. Punchy expressive face, quick gestures. Vertical 9:16 framing, waist-up composition, no on-screen text.

[SHOT] Vertical 9:16, 1080x1920, 30fps. Full-frame host shot, chest-up composition, eyes on the lens, head centred with breathing room above.

[BACKGROUND] Locked to the registered studio backdrop supplied as background.png. Do not redress, relight or replace the set — every segment of this project must cut together.

[LIGHTING] Soft key from camera-left, gentle rim separation from the backdrop, constant colour temperature and exposure across all segments.

[ACTION] Opening hook: already speaking at frame 1 — no dead air, no intro beat. Energy starts high and settles by the second sentence. Delivery follows the host's registered register (活潑、節奏快的迷因口吻；愛用網路用語，但事實與風險必說).

[LIP-SYNC] Drive the mouth from the supplied audio.wav (15.0s of Mandarin narration). Mouth stops the moment the audio stops.

[DURATION] 15.0s, one continuous take, no cuts.

[IDENTITY] Match host_reference.png exactly — same person, hair, wardrobe and framing as every other segment of project proj_04. This project is locked to 周曉薇 (@meme_coin_diary).

由主持人的 base_prompt 加上本段語境（版型、開場／中段／CTA、對嘴長度、專案一致性）組出， 不需要人工再編輯。

Negative Prompt畫面不得生成字幕spec 第 6 節：字幕由 Ditto-Vid 在最終合成時另外加，Higgsfield 畫面本身不得生成字幕。

on-screen text, subtitles, captions, burned-in captions, karaoke text, lower third, watermark, logo, channel branding, timecode overlay, second person in frame, crowd, background actors, camera cut, jump cut, hard zoom, whip pan, shaky handheld, background change, set change, wardrobe change mid-shot, extra fingers, deformed hands, warped facial features, identity drift, lip-sync drift, mouth moving after audio ends, cartoon style, anime style, plastic skin, heavy beauty filter, motion blur artifacts, flicker, glitch, frame duplication, any visual suggestion of: 保證獲利, 穩賺, 無風險, 一定漲, 內線消息, 報明牌, 全押, 梭哈

任務包內容

  • task.json

    任務參數：prompt、negative_prompt、target_duration、輸出檔名

    project_id=proj_04｜segment_id=seg_0401｜host_id=host_04

  • audio.wav

    該段 Cartesia 音檔，作為 Higgsfield 的對嘴音軌

    尚未生成 — 音檔就緒前無法下載任務包

  • host_reference.png

    主持人參考圖，確保十位主持人不會互相串味

    周曉薇（host_04）的固定形象，Demo 以漸層色塊代替真實圖檔

  • background.png

    該主持人的固定背景，讓所有片段能剪在一起

    同一專案所有片段共用，不得在 Higgsfield 端替換

  • instructions.txt

    操作步驟、輸出檔名與 QA 檢查清單

    輸出檔名 proj_04_seg_0401_v1.mp4｜目標 15.0 秒

段 3seg_0403原片＋主持人 PIP草稿待生成spec 版本 A 第 7 節：每段狀態顯示待生成 / 已上傳 / 待審核 / 核准 / 退回重生
音檔尚未就緒：請先到「音檔生成」分頁完成本段 Cartesia 音檔，任務包才會有對嘴音軌。
host_reference.png
background.png

周曉薇

@meme_coin_diary
ctv_zh_tw_female_meme_04

音檔尚未就緒：請先到「音檔生成」分頁完成本段 Cartesia 音檔，任務包才會有對嘴音軌。

Higgsfield Prompt16.2sproj_04_seg_0403_v1.mp4

[HOST] A playful Taiwanese female creator in her mid-20s, high ponytail, oversized graphic tee, bouncing slightly in a neon-lit bedroom studio with LED strips. Punchy expressive face, quick gestures. Vertical 9:16 framing, waist-up composition, no on-screen text.

[SHOT] Vertical 9:16, 1080x1920, 30fps. Tight picture-in-picture crop — the host is composited at 28% frame width in the bottom-right corner with a 4% margin, so keep the head centred in the upper third of the plate and leave clean space around it.

[BACKGROUND] Locked to the registered studio backdrop supplied as background.png. Do not redress, relight or replace the set — every segment of this project must cut together.

[LIGHTING] Soft key from camera-left, gentle rim separation from the backdrop, constant colour temperature and exposure across all segments.

[ACTION] Mid-roll explanation: measured delivery, small purposeful gestures on key numbers, no large body movement that would break the composite. Delivery follows the host's registered register (活潑、節奏快的迷因口吻；愛用網路用語，但事實與風險必說).

[LIP-SYNC] Drive the mouth from the supplied audio.wav (16.2s of Mandarin narration). Mouth stops the moment the audio stops.

[DURATION] 16.2s, one continuous take, no cuts.

[IDENTITY] Match host_reference.png exactly — same person, hair, wardrobe and framing as every other segment of project proj_04. This project is locked to 周曉薇 (@meme_coin_diary).

由主持人的 base_prompt 加上本段語境（版型、開場／中段／CTA、對嘴長度、專案一致性）組出， 不需要人工再編輯。

Negative Prompt畫面不得生成字幕spec 第 6 節：字幕由 Ditto-Vid 在最終合成時另外加，Higgsfield 畫面本身不得生成字幕。

on-screen text, subtitles, captions, burned-in captions, karaoke text, lower third, watermark, logo, channel branding, timecode overlay, second person in frame, crowd, background actors, camera cut, jump cut, hard zoom, whip pan, shaky handheld, background change, set change, wardrobe change mid-shot, extra fingers, deformed hands, warped facial features, identity drift, lip-sync drift, mouth moving after audio ends, cartoon style, anime style, plastic skin, heavy beauty filter, motion blur artifacts, flicker, glitch, frame duplication, any visual suggestion of: 保證獲利, 穩賺, 無風險, 一定漲, 內線消息, 報明牌, 全押, 梭哈

任務包內容

  • task.json

    任務參數：prompt、negative_prompt、target_duration、輸出檔名

    project_id=proj_04｜segment_id=seg_0403｜host_id=host_04

  • audio.wav

    該段 Cartesia 音檔，作為 Higgsfield 的對嘴音軌

    尚未生成 — 音檔就緒前無法下載任務包

  • host_reference.png

    主持人參考圖，確保十位主持人不會互相串味

    周曉薇（host_04）的固定形象，Demo 以漸層色塊代替真實圖檔

  • background.png

    該主持人的固定背景，讓所有片段能剪在一起

    同一專案所有片段共用，不得在 Higgsfield 端替換

  • instructions.txt

    操作步驟、輸出檔名與 QA 檢查清單

    輸出檔名 proj_04_seg_0403_v1.mp4｜目標 16.2 秒

段 4seg_0404主持人全畫面已驗證待生成spec 版本 A 第 7 節：每段狀態顯示待生成 / 已上傳 / 待審核 / 核准 / 退回重生
音檔尚未就緒：請先到「音檔生成」分頁完成本段 Cartesia 音檔，任務包才會有對嘴音軌。
host_reference.png
background.png

周曉薇

@meme_coin_diary
ctv_zh_tw_female_meme_04

音檔尚未就緒：請先到「音檔生成」分頁完成本段 Cartesia 音檔，任務包才會有對嘴音軌。

Higgsfield Prompt14.4sproj_04_seg_0404_v1.mp4

[HOST] A playful Taiwanese female creator in her mid-20s, high ponytail, oversized graphic tee, bouncing slightly in a neon-lit bedroom studio with LED strips. Punchy expressive face, quick gestures. Vertical 9:16 framing, waist-up composition, no on-screen text.

[SHOT] Vertical 9:16, 1080x1920, 30fps. Full-frame host shot, chest-up composition, eyes on the lens, head centred with breathing room above.

[BACKGROUND] Locked to the registered studio backdrop supplied as background.png. Do not redress, relight or replace the set — every segment of this project must cut together.

[LIGHTING] Soft key from camera-left, gentle rim separation from the backdrop, constant colour temperature and exposure across all segments.

[ACTION] Closing call-to-action: settle the pace, lean in slightly on the final sentence, hold a calm confident expression for the last 0.5s without moving out of frame. Delivery follows the host's registered register (活潑、節奏快的迷因口吻；愛用網路用語，但事實與風險必說).

[LIP-SYNC] Drive the mouth from the supplied audio.wav (14.4s of Mandarin narration). Mouth stops the moment the audio stops.

[DURATION] 14.4s, one continuous take, no cuts.

[IDENTITY] Match host_reference.png exactly — same person, hair, wardrobe and framing as every other segment of project proj_04. This project is locked to 周曉薇 (@meme_coin_diary).

由主持人的 base_prompt 加上本段語境（版型、開場／中段／CTA、對嘴長度、專案一致性）組出， 不需要人工再編輯。

Negative Prompt畫面不得生成字幕spec 第 6 節：字幕由 Ditto-Vid 在最終合成時另外加，Higgsfield 畫面本身不得生成字幕。

on-screen text, subtitles, captions, burned-in captions, karaoke text, lower third, watermark, logo, channel branding, timecode overlay, second person in frame, crowd, background actors, camera cut, jump cut, hard zoom, whip pan, shaky handheld, background change, set change, wardrobe change mid-shot, extra fingers, deformed hands, warped facial features, identity drift, lip-sync drift, mouth moving after audio ends, cartoon style, anime style, plastic skin, heavy beauty filter, motion blur artifacts, flicker, glitch, frame duplication, any visual suggestion of: 保證獲利, 穩賺, 無風險, 一定漲, 內線消息, 報明牌, 全押, 梭哈

任務包內容

  • task.json

    任務參數：prompt、negative_prompt、target_duration、輸出檔名

    project_id=proj_04｜segment_id=seg_0404｜host_id=host_04

  • audio.wav

    該段 Cartesia 音檔，作為 Higgsfield 的對嘴音軌

    尚未生成 — 音檔就緒前無法下載任務包

  • host_reference.png

    主持人參考圖，確保十位主持人不會互相串味

    周曉薇（host_04）的固定形象，Demo 以漸層色塊代替真實圖檔

  • background.png

    該主持人的固定背景，讓所有片段能剪在一起

    同一專案所有片段共用，不得在 Higgsfield 端替換

  • instructions.txt

    操作步驟、輸出檔名與 QA 檢查清單

    輸出檔名 proj_04_seg_0404_v1.mp4｜目標 14.4 秒

不需要主持人動畫的段落

原片滿版段落直接使用 YouTube 原素材與原音，不進 Higgsfield 流程。

  • 段 2seg_0402原片滿版已驗證Higgsfield 不適用