female-outfit-director: A Codex Skill for Generating AI Outfit Change Video Prompts

What female-outfit-director Does and Who It Is For

Content creators who produce outfit change videos with AI image and video generation tools face a structuring problem. Generating consistent character appearance across five different outfits, with smooth visual transitions, requires careful prompt engineering that accounts for character identity, garment design, layout, camera position, and audio timing. Doing this freehand in a conversational AI session produces inconsistent results. female-outfit-director addresses this by encoding the entire production schema as a Codex skill. When invoked, the skill asks for a reference image or character description and then outputs four artifacts: a set of parameter confirmations showing the locked settings, five collage frame image prompts with consistent character features across all looks, a per-segment video timeline with precise beat points and transition instructions, and a negative prompt list to prevent common generation errors. The skill targets adult female characters by default. If no reference image is given and no nationality is specified, the skill uses an original adult Chinese female character design. Users who work with a reference image should use a model with image understanding capability.

Default Format: 9:16, 8 Seconds, Five Outfits, Fixed Camera

The skill ships with a set of default parameters that define the production format. The default aspect ratio is 9:16 for vertical short video. Duration is 8 seconds. The number of looks is five: an initial outfit plus four outfit changes. The composition places one large central character figure with four floating full-body character stickers in the corners. The camera is fixed. Audio defaults to no background music, using only click, fly-in, and beat sound effects at transition points; if the user specifies music, the skill schedules it. Every parameter can be overridden. The README gives examples of requesting a 15-second dance video, a different number of characters, different camera angles, or custom beat timings. This override design means the defaults cover the most common vertical short video format while allowing complete customization without rewriting the skill.

Installing female-outfit-director on Windows with Codex

The installation path is documented for Windows using PowerShell. The skill directory goes into the Codex global skills location:

git clone https://github.com/liyue-aigc/female-outfit-director.git
New-Item -ItemType Directory -Force "$HOME\.codex\skills\female-outfit-director"
Copy-Item -Recurse -Force ".\female-outfit-director\skill\*" "$HOME\.codex\skills\female-outfit-director\"

After installation, restart the Codex task session. The skill is then available with the prefix:

$female-outfit-director 帮我为同一位中国成年女性设计五套都市穿搭拼贴首帧，并生成8秒卡点换装视频脚本。

The full installation guide with additional steps is in docs/INSTALLATION.md, and a first-use tutorial is in docs/FIRST_USE.md. The README recommends reading the first-use tutorial before running the skill for the first time.

The M1 Composite Transition Mechanism

The default transition mechanism is called M1, described as a consumable sticker swap. The README documents the exact sequence of steps. First, the user taps one of the four floating character stickers. The skill creates a copy of that sticker as a fly-in duplicate. The original sticker position immediately empties, leaving a gap. The copy moves toward the center while enlarging. The copy aligns with the central character at head, shoulders, waist, and pose keypoints, triggering the outfit change. Both the fly-in copy and a white dashed outline disappear, leaving a single changed character at center with no ghost or double-character artifact. This mechanism ensures that after each transition the small character count decreases by one. The skill includes 11 additional transition mechanisms covering character sticker fly-in variants, sleeve-cover transitions, spinning relay, sticker page-flip, and action beat-point transitions. The user can specify any mechanism or let the skill select the default M1 for each change.

What the Skill Does Not Provide and Its Limitations

The skill produces text prompts only. It does not call any image generation or video generation service. After running the skill, the user still needs to take the output prompts and submit them to a separate generation tool such as a diffusion model or a video synthesis platform. The README states explicitly that the repository does not include third-party image or video generation services. The skill is written in Chinese, and the invocation command in the README is in Simplified Chinese. English speakers can install and run the skill but the output will be in Chinese unless they override the language in their prompt. The compatibility list covers only Codex and agents that support local Skill files. The README does not document support for Claude Code, Gemini CLI, or other agents.

Compared to Generic Image Prompt Templates and Licence Terms

The main alternative is writing outfit change prompts manually or using a generic prompt template library. A manual approach gives full control but requires the author to remember all consistency constraints (character identity, layout grid, transition beat timing, negative prompts) on each session. A generic template library such as a Stable Diffusion prompt collection covers many image styles but does not encode the specific structured format of a multi-look outfit change video with composite sticker mechanics. female-outfit-director encodes the domain knowledge of the outfit change short video genre specifically, including the five-look collage layout, the beat-point timing structure, and the 12 named transition types. The trade-off is that the skill is narrow by design; it does not generalize to other video types without modification. The project is licensed under MIT, which permits commercial use and modification without restrictions on redistribution.

Editorial conclusion

female-outfit-director is the right tool for content creators and AI video producers who want a repeatable, parameter-driven workflow for outfit change clips without writing raw generation prompts from scratch each time. It covers the full production chain from collage frame prompt to per-segment video script with sound cues. Users who need a service that actually runs the image or video generation will need to supply their own generation pipeline; the skill produces text prompts only. The MIT licence permits commercial use and modification. The repository does not include a Docker image or a web interface; it is a plain text skill file installed into a local Codex environment. The last push was on 2026-09-07.

Frequently asked questions

Official sources

1. Issues
2. License: MIT
3. liyue-aigc/female-outfit-director on GitHub
4. README