AUTONOMOUS SCRIPT → SCREEN · LONG-HORIZON TOOL AGENT

WeVisCut

The Whole Editing Room in One Agent
From Narration to Long-Form Video

Chenyu Mu^1,2, Wenming Weng², Qinrui Wang², Xingsong Ye², Hao Yu², Jun He², Chong Sun^2,✉, Chen Li², Jing LYU², Erkun Yang^1,✉, Cheng Deng¹

¹Xidian University ²WeChat Vision, Tencent Inc. ✉ corresponding authors

WeVisCut connects tool execution with policy learning for long-form script-to-video production. Editing inversion and verified replay recover executable supervision from 3,750 finished projects, yielding 134,771 training examples for a single 9B policy. On 100 held-out tasks, it achieves a 90.5 overall score and uses about 33× fewer generated tokens than Claude Opus 5 per successful output minute.

MODEL DATA-Pipeline TECH REPORT

WATCH IT EDIT RESULTS GALLERY

DEMO · LIVE RUN

Watch the Agent Cut — Then Tell It What to Change

Every turn is observe → think → act. After the first cut lands, the director chats: swap a shot's source, change the music, restyle the captions — the policy re-plans locally and re-delivers, never re-shooting what already worked.

REC WeVisCut · AGENT DECK TAKE A7F2 00:00:00:00

DIRECTOR'S CHAT · 与智能体对话

PREVIEW · 成片预览

AWAITING FIRST CUT

CUT v1

TIMELINE · 时间线实时构建

TXT

BGM

segments–

shots placed–

open issues0

repair budget4

artifact–

Scripted replay of real production + interactive-revision traces · AICut Studio runtime · DAPO checkpoint

DATA ENGINE

Editing Inversion & Verified Replay

We recover editing intent from finished narration-driven videos, then replay it with production tools. The four-stage pipeline adds observation, bounded repair and trajectory materialization to produce executable supervision.

Four-stage executable data construction pipeline — FIG.3 — EDITING INVERSION → VERIFIED REPLAY → OBSERVE–REFLECT–REPAIR → TRAJECTORY MATERIALIZATION

0finished video projects

0training examples

0held-out evaluation tasks

METHOD

One Shared Policy, Stateful Tool Execution

WeVisCut maintains shared production state across tool calls. Runtime feedback informs subsequent decisions, inspection, bounded repair and replanning instead of following a fixed open-loop workflow.

WeVisCut system overview from script planning through video delivery — FIG.1 — WEVISCUT OVERVIEW · PLANNING · RETRIEVAL · COMPOSITION · INSPECTION · REPAIR

WeVisCut framework linking data construction, policy training and inference — WEVISCUT FRAMEWORK — DATA CONSTRUCTION → TRAINING → INFERENCE · CLOSED-LOOP TOOL EXECUTION

Orchestrator

Selects legal tools from the current production state and coordinates planning, execution, inspection and delivery.

Generator

Creates scripts and shot plans, retrieves visual and audio assets, and composes narration, decorations and timeline edits.

Critic

Inspects returned artifacts against quality dimensions, identifies failures and routes targeted repair.

Shared Production State

Validated tool returns update canonical state and artifacts, so each next action uses the latest run status.

↻

Closed-Loop Repair

Local fixes preserve successful work; inspection evidence feeds back into the policy for the next decision.

GALLERY

Directed by the Policy

Unedited deliveries from the deployed runtime. One script in — one finished film out.

▶03:21

To Space, On a Rocket

太空科普 · 火箭如何送人上天

▶05:49

Five Thousand Years

中华文明 · 时间洪流中的延续

▶04:37

How Trees Lift Water

没有水泵 · 大树如何送水上树冠

RESULTS

Small Policy, Frontier Delivery

100 held-out script-to-video tasks · 5 independent runs · 5 metrics (mean ± std), descriptive Overall score and cost per successful output minute.

METHOD	SCF↑	VNSM↑	TPR↑	LHC↑	ERY↑	OVERALL↑	CALLS/SUCCESS MIN↓	TOKENS/SUCCESS MIN↓
GENERAL-PURPOSE BASELINES
Qwen3.5-9B	69.8±1.6	59.2±1.8	79.5±1.7	93.7±0.8	70.4±2.1	74.5	3.1	11,883
Qwen3.6-27B	72.6±1.4	59.7±1.6	80.1±1.5	95.1±0.7	72.1±1.9	75.9	3.9	6,742
Qwen3.6-35B-A3B	71.9±1.9	56.7±2.0	83.4±1.8	97.3±0.6	79.4±2.3	77.7	4.7	27,554
REFERENCE MODELS
Qwen3.5-397B-A17B	75.8±1.5	64.1±1.7	84.6±1.6	96.2±0.8	86.9±1.8	81.5	6.0	15,223
Gemini 3.1 Pro	80.2±1.2	66.5±1.3	88.2±1.2	98.0±0.5	90.3±1.5	84.6	4.5	13,814
Claude Opus 5	87.1±1.0	74.8±1.1	93.1±1.0	98.6±0.4	95.8±1.2	89.9	11.6	78,763
OUR METHOD
WeVisCut (Ours)	88.4±2.1	74.6±2.3	94.0±2.0	98.8±0.5	96.7±1.8	90.5	3.1	2,377

Common-harness results on 100 held-out tasks across five runs; Overall is the descriptive mean of the five task metrics.

NATIVE SYSTEM	SCF↑	VNSM↑	TPR↑	LHC↑	ERY↑	OVERALL↑	WALL MIN/OUTPUT MIN↓	USD/OUTPUT MIN↓
Crayotter	81.6	67.8	89.1	94.2	91.6	84.9	8.6	0.97
VideoAgent	78.4	65.2	86.7	93.5	88.7	82.5	10.4	0.58
WeVisCut	88.4	74.6	94.0	98.8	96.7	90.5	2.1	0.09

Task-matched native systems use the same script, source pool, duration and aspect ratio. Cost includes unsuccessful attempts.

HUMAN MOS (1–5)	THEME	RICH	NARR	SMOOTH	VISUAL	OVERALL
Crayotter	3.58	3.42	3.35	3.48	3.55	3.48
VideoAgent	3.46	3.60	3.30	3.42	3.52	3.46
WeVisCut	4.45	4.32	4.58	4.52	4.38	4.56

Blinded human evaluation on 100 held-out cases · 3 qualified raters · descriptive mean ratings.

Training progression on SCF, VNSM, LHC — FIG.5 — TRAINING PROGRESSION ON SCF, VNSM AND LHC · GUIDED-OPD → RCF-DAPO

READING THE NUMBERS

①

Scaling alone does not specialize. The Qwen3.6-35B-A3B baseline reaches 77.7 Overall, Gemini 3.1 Pro 84.6, and WeVisCut 90.5. Task-specific post-training closes the gap beyond model scale alone.

②

On par with Claude Opus 5 — at 33.1× fewer tokens. Better on SCF (+1.3), TPR (+0.9), LHC (+0.2), ERY (+0.9), −0.2 on VNSM; every metric within 1.3 points, with 97.0% fewer generated tokens and 73.3% fewer calls per output minute.

③

Native-stack advantage holds end-to-end. +6.8 SCF over Crayotter and +10.0 over VideoAgent, while running 4.1× / 5.0× faster at 10.8× / 6.4× lower cost — and humans prefer it on every dimension (4.56 vs 3.48 overall).

④

Three training phases specialize the 9B policy. Unified SFT and runtime-verified RFT initialize tool use and recovery; Guided-OPD trains recovery decisions; RCF-DAPO optimizes decision quality. Overall rises from 74.5 for the base model to 90.5.

EVALUATION SANDBOX

Scores With Evidence

A finished film is sampled uniformly per shot; the harness confirms keyframes as visual evidence and runs its analysis toolchain — layout, saliency, legibility, motion-overlap — before the held-out Claude Opus 5 judge scores two rubric dimensions: visual_impact and clarity , each with detailed reasoning and cited frames.

EVAL FROZEN HARNESS · v2.3 CASE #0042 00:00:00:00

INPUT · 被评影片

final_cut.mp4 · 16:58 · 21 shots

UNIFORM SHOT SAMPLING · 逐镜头均匀抽帧

SANDBOX · HELD-OUT JUDGE · CLAUDE OPUS 5

VISUAL_IMPACT

视觉表现力 · 布局 / 美观度 / 吸引力

8.8/10

构图遵循三分法，主体稳定处于视觉重心；暖棕复古色调全片统一，特写与中景交替避免视觉疲劳，转场节奏贴合旁白呼吸感。扣分项：S11 叠化时长略拖。

CLARITY

信息清晰度 · 文字可读 / 动效不干扰

9.2/10

字幕与背景对比度 7.2:1，全片无文字溢出与遮挡；关键信息均与图示动画同步出现，动效未干扰正文阅读；S07 信息密度偏高，但仍在可读阈值之内。

Auditable VNSM evaluator path — FIG.6 — AUDITABLE VNSM EVALUATION · SHOT PACKET → 4 TIMESTAMPED FRAMES (OPTIONAL BOUNDED CLIP) → VISIBLE EVIDENCE → ANCHORED GRADE + VALIDATED JSON

HEAD-TO-HEAD

One Script, Two Cuts

Four held-out scripts, each produced by both systems from the same script, source pool, duration and aspect ratio. Pick a domain — both sides switch together; press play, both rolls run in sync.

WeVisCut · OURS

SYNCED

CRAYOTTER

SCRIPT · 人工智能如何改变生活

CITE

Citation

BIBTEX

@misc{mu2026weviscut,
  title  = {WeVisCut: The Whole Editing Room in One Agent: From Narration to Long-Form Video},
  author = {Mu, Chenyu and Weng, Wenming and Wang, Qinrui and Ye, Xingsong and Yu, Hao and He, Jun and Sun, Chong and Li, Chen and LYU, Jing and Yang, Erkun and Deng, Cheng},
  year   = {2026}
}