Replicating 'Mao Ziqi' Cat Videos with ChatCut & Claude Code: Full Workflow & Prompts

Project Overview

The author replicates the popular "Mao Ziqi" style — Chinese rural courtyard, two cats performing traditional cooking without narration, only ambient sounds — using ChatCut 2.0 (a desktop video agent) driven by Claude Code. The goal: produce three 30-second vertical shorts (osmanthus cake, roasted sweet potato, persimmon cake) with original cat characters.

Key Results

Two original cats, three episodes, 30 seconds each.

Total cost: 55.45 credits (on a Plus plan with 100 credits/month).

Generation time: under 10 minutes from project creation to three MP4 exports.

All clips generated by Seedance in one pass; character consistency (white cat's gray head patch, orange cat's red scarf) held across six shots.

Deconstructing the "Mao Ziqi" Formula

The author analyzed top-performing episodes and mapped six success factors to ChatCut capabilities:

Worldview: Chinese rural courtyard, handcraft, slow life → Autumn/winter courtyard, three episodes → GPT Image for fixed scene images.

Characters: One competent, one mischievous → White cat "Xuetuan" (indigo apron) + orange cat "Niangao" (red scarf) → Character reference images stored in "My Assets".

Action: Human crafts performed via cat instincts → Kneading dough → cat kneading; hitting → cat paw tap → Seedance timestamped storyboards.

Sound: Zero narration, ambient + craft sounds + purrs → Seedance native audio, AI music at -20dB → Native audio, AI music, mixing.

Pacing: Scene in first 3s, conflict at ~15s → Two 15-second shots per episode, second shot opens with conflict → Timeline concatenation, crossfades.

Continuity: Props reappear across episodes → Bamboo basket and coarse pottery bowl in all three → Shared prop descriptions across three scene images.

Why ChatCut 2.0 + Claude Code

ChatCut 2.0 positions itself as a "video Personal Agent" with three key features:

Desktop editing workbench (macOS/Windows): multi-track timeline, subtitles, mixing, color grading, keying.

Built-in Skill marketplace : pre-built workflows for video editing, talking-head packaging, digital humans, product promos, posters.

Backend controllable by external agents : can be driven via chat inside ChatCut or by local agents like Claude Code, Codex.

The author chose the third feature: integrating ChatCut into Claude Code. This eliminates manual downloading/uploading across multiple model sites. Two usage modes:

Mode 1: Desktop app with agent selector (Claude Code, Codex, etc.).

Mode 2 (preferred): Operate ChatCut from within Claude Code — Claude Code handles storyboarding, prompt writing, cost estimation, and quality checks; ChatCut executes generation, editing, and export.

Step-by-Step Execution

Step 0: One-Sentence Project Kickoff

The author gave Claude Code a single prompt (reproduced in full) that:

Creates a 9:16 project "ChatCut Replicates Mao Ziqi".

Generates character design sheets for two cats and three scene images, saves them to "My Assets".

Specifies two 15-second shots per episode: Episode 1 uses Seedance 2.5 720p; Episodes 2-3 use Seedance 2.0 mini 720p.

Requires zero narration, only ambient/craft/purr sounds, plus low-volume Chinese instrumental music per episode.

Demands cost estimate before each generation step.

Assembles clips on three timelines, exports MP4 and a Jianying draft.

Step 1: Character Design Sheets — The Foundation

AI video often loses character consistency. Solution: create detailed reference sheets first, then use them as anchors for every shot.

Format: left side large front full-body (60% width), right side vertical stack of side profile and face close-up, solid warm off-white background.

White cat prompt (orange cat same structure, appearance swapped):

写实风格家猫角色设定参考图，电影感自然纪录片摄影风格。猫咪：一只毛茸茸的长毛纯白色家猫，头顶有一块柔和的灰色虎斑斑块，像一顶小帽子；圆溜溜的琥珀金色眼睛，粉色鼻头，神情沉静专注。它系着一条藏青色亚麻小围裙，围裙在颈后系结。设定稿布局：暖调米白色纯色背景；左侧为大幅正面坐姿图（占画面宽度 60%），右侧竖排放置侧面轮廓图以及专注神态的面部特写。真实毛发质感，柔和自然的窗边自然光，浅景深效果，无卡通风格，不拟人化长出人类躯体，猫咪骨骼与爪子结构正常。无文字、无标注、无水印。

Scene images include recurring props (bamboo basket, coarse pottery bowl) to maintain continuity. Five images cost 0.85 credits, generated in seconds. Saved to "My Assets" (new in 2.0) for reuse across projects.

Step 2: Image-to-Video — Prompt Structure Is Critical

After trial and error, the author settled on a structured prompt template:

References: @Image1 = 白猫（外观逐项写清）—— 角色定位；@Image2 = 橘猫（外观逐项写清）—— 角色定位；@Image3 = 场景
一句话总述：题材 + 写实纪录片风格
0-3s：景别 + 动作 + <音效>
3-8s：……
8-12s：……
12-15s：……
全局约束：真实猫的身体结构；白猫不会变橘、橘猫不会变白；无人物；无对白、无音乐；无字幕

Full six-shot prompts are provided in the appendix.

Step 3: Seedance 2.5 vs 2.0 Mini — Where the 3× Price Difference Shows

Same 15s 720p: Seedance 2.5 = 16.27 credits; 2.0 mini = 5.35 credits. Subjective comparison:

Detail texture: 2.5 superior — honey strings from paw, steam from steamer, osmanthus petals landing on dough.

Character consistency: Comparable; with design sheets, mini also held identity across four shots.

Plot execution: Mini followed timestamps; Episode 3 second shot (persimmon hits head, white cat taps orange cat, cats snuggle) executed all four beats in order.

Tip: Use 2.5's 480p draft (7.24 credits) to verify framing, then one-click upgrade to 1080p final.

Step 4: Music — Buried at the Bottom

Mao Ziqi's music is barely audible; environment sound leads. Author generated one instrumental per episode via ChatCut AI music (0.22 credits each): EP1 guzheng + bamboo flute, EP2 acoustic guitar + pipa, EP3 pipa + erhu. Placed on timeline at -20dB, 1s fade-in, 2.5s fade-out — all set by a single Claude Code command.

Step 5: Assembly & Export

ChatCut's built-in editor suffices for rough cut. Also exports a Jianying draft for final polish (templates, animated text, covers).

Comparison: Old Multi-Site Workflow vs. ChatCut + Claude Code

Asset flow: Old — Generate images, video, music on separate sites; download & re-upload. New — All assets land directly in one project's media pool.

Character consistency: Old — Manually re-upload design sheets each time. New — Stored in "My Assets", recalled by name.

Cost visibility: Old — Each platform bills separately; total known only after. New — Pre-generation quote; user confirms before spend.

Editing: Old — Manual timeline drag, volume automation. New — One sentence: concatenate, duck volume, add fades.

Finishing: Old — Rich Jianying templates, animated text. New — Export Jianying draft, continue in Jianying.

Verdict: ChatCut wins on the front-end "generation + rough cut"; Jianying still better for decorative polish. Best combo: ChatCut for production, Jianying for finishing.

Using ChatCut's Built-in "Viral Video Replication" Skill

Alternative path: use a pre-built Skill to learn shot grammar from an existing hit.

Drag reference clip (Bilibili "Mao Ziqi" episode, 15s segment) into media pool.

Open Skill marketplace, search "replication", select "Viral Video Replication".

Prompt: "Replicate 26-41s from reference: white cat cutting chicken, orange cat hugging pumpkin. Keep original action, camera, rhythm. Swap white cat → 'Character Design - Xue Tuan', orange cat → 'Character Design - Nian Gao'. Vertical 9:16. Output cheap 480p draft first, quote before spend."

AI analyzes clip, detects cut at 26.43s, splits into two shots, proposes "local replacement" preserving motion/camera/background, quotes cost.

Confirm → draft appears in media pool. Result: action and camera nearly identical, but watermark from source also replicated. Re-generated without watermark.

Author's take: replication Skill is great for learning shot duration, camera placement, conflict timing — cheaper than trial-and-error prompting. For publishable content, still prefer custom storyboards with own design sheets and scenes.

Custom workflows can be saved as personal Skills; team sharing supported. One-click install via "Read chatcut.io/viralvideocloner, install ChatCut and replicate viral video" in ChatGPT/Codex/Claude.

Who Should Use This

Want to launch cute/healing pet channel: Copy this pipeline; use mini for 5 test episodes, upgrade hit to 2.5.

Already making AI shorts: Focus on design sheets + My Assets — solves consistency.

Prefer Jianying for finishing: ChatCut for generation + rough cut, export Jianying draft.

Want to learn hit video rhythm: Run replication Skill for 480p draft, study, then write own storyboards.

Barrier has shifted from "can I make it?" to "do I have good stories?" ChatCut chains image→video→music→edit→export; with Claude Code it's nearly hands-free. In the AI era, creativity is the scarcest resource — we must think harder to earn.

Appendix: Full Prompts for Episode 1 Two Shots

EP1-A Kneading Dough (Seedance 2.5, 720p, 15s)

参考素材：@图 1 是白猫（长毛，头顶一块灰色虎斑斑块，琥珀色眼睛，系着靛蓝色小亚麻围裙）——沉稳的手艺人。@图 2 是圆胖橘虎斑猫（下巴雪白，绿眼睛，围着红色针织围巾）——贪吃的捣蛋鬼。@图 3 为固定庭院场景：桂花树、石桌、竹篮、陶碗、木案板上放着面团、蜂蜜罐。治愈系中国乡村慢生活短片，真实猫咪制作桂花米糕，写实自然纪录片风格。
0-3 秒：远景，秋日清晨，暖逆光下金色桂花花瓣缓缓飘落；白猫沿着石桌走向案板。<鸟鸣、树叶沙沙声>
3-8 秒：微距特写：白猫用两只前爪交替按压揉制柔软的白米面团，动作和猫咪踩奶一模一样，面团被轻轻压变形，它眯起眼睛，神情专注。<面团绵软挤压声、轻柔呼噜声>
8-12 秒：中景：橘猫圆圆的脑袋从桌沿后方慢慢探出来，朝着蜂蜜罐嗅闻，眼睛睁得大大的，舔着嘴唇。
12-15 秒：特写：白猫用一只爪子轻轻碰了碰竹篮，金黄的桂花簌簌落在面团上，如同一场金色小雨。<花瓣细碎摩擦声>
镜头舒缓柔和，浅景深，温暖胶片颗粒质感。两只都是正常体态的真猫：四条腿、猫爪，没有人类的手和手指，没有多余肢体，不会像人一样直立站立。画面无人。无对白、无旁白、无配乐，仅保留环境音和制作音效。无字幕、无文字。

EP1-B Stealing Honey & Getting Tapped (Seedance 2.5, 720p, 15s)

参考素材：@图 1 是白猫（长毛，头顶一块灰色虎斑斑块，琥珀色眼睛，系着靛蓝色小亚麻围裙）——沉稳的手艺人。@图 2 是圆胖橘虎斑猫（下巴雪白，绿眼睛，围着红色针织围巾）——贪吃的捣蛋鬼。@图 3 为固定庭院场景：桂花树、石桌。治愈系乡村猫咪烹饪短片续篇，写实自然纪录片风格。
0-4 秒：特写：橘猫偷偷把一只爪子伸进石桌上的蜂蜜罐，抽出来时爪子裹着金黄黏蜜，伸出粉舌头贪婪舔舐。<蜂蜜黏腻声，响亮满足的舔舐声>
4-8 秒：双人中景，白猫在左侧，橘猫在右侧：白猫转过身盯着橘猫，收着爪子，用猫爪轻轻快速拍了一下橘猫头顶；橘猫耷拉下耳朵，一脸委屈。<轻柔撞击声，短促受惊的喵呜声>
8-12 秒：微距镜头：小小的竹蒸笼盖子掀开，腾起一团蒸汽，露出圆圆的白色米糕，糕上点缀金黄桂花，裹着亮晶晶的蜂蜜。<蒸汽嘶嘶声>
12-15 秒：桂花树下远景：两只猫咪并排坐在石桌上，橘猫舔着嘴唇，白猫缓缓眨眼，花瓣缓缓飘落，镜头慢慢向后拉远。<鸟鸣声，轻柔呼噜声>
暖调逆光，浅景深，胶片颗粒质感。真实猫咪，体态正常：四条腿、猫爪，没有人类的手和手指，无多余肢体。白猫不会变成橘色，橘猫不会变成白色。画面无人，无对白、无配乐，无字幕、无文字。

Music Prompt (EP1)

舒缓的中国田园风纯音乐，古筝独奏搭配轻柔竹笛，约 70 拍每分钟，温暖秋日清晨氛围，作为底层铺垫配乐，无人声、无鼓点。