What does Clothing Promo Video do?
Use this skill's built-in parametrized renderer to generate real MP4 files. The model is responsible for analyzing images, classifying them, and orchestrating the composition; the script handles animation, music, encoding, and full decoding verification. Runs in any working directory, no need for the original web project, API keys, or project-specific tools.
Assets and Facts
- Actually inspect the images provided by the user this time; distinguish between main clothing images, detail shots, duplicates, brand images, and reference images. Reference images are only for comparison and will not appear in the final video; record the classification for every image, and explain why any image is not shown.
- By default, preserve the original appearance and aspect ratio of clothing and people; morphing is applied only to the display container. Cropping is only done after visually confirming the subject is safe, and the cropping boundary and reason must be recorded.
- This style uses LOOK numbers and fixed outfit copy; do not infer colors, fabric composition, functionality, price, discounts, inventory, contact information, or brand authorization from images. Do not treat LOOK numbers as real product codes. Only brand images explicitly specified by the user may be used for the brand ending; do not extract store branding from logos on clothing or old assets.
- Supports 1–12 looks and at most one brand image for the ending. By default, every identified look is shown; if the video length is clearly insufficient, explain the required duration and ask the user to either extend or group the looks—do not discard any look arbitrarily. If only a brand image is provided, supplement it with a main clothing image.
- When the user explicitly requests custom titles, ad copy, brand-new effects, or realistic runway simulations, explain the actual scope of this renderer; do not claim unimplemented parameters as realized. Different aspect ratios are handled by centering vertically and cropping equally, preserving portrait composition.
Execution
Replace SKILL_DIR below with the actual skill directory path as read, and PYTHON with a Python interpreter that has the necessary dependencies; do not hardcode the original project path.
-
Run
PYTHON SKILL_DIR/scripts/render_video.py --check. An existing Python environment can be reused; if dependencies are missing, create a virtual environment in the current working directory and runpython -m pip install -r SKILL_DIR/scripts/requirements.txt. Dependencies are Pillow, NumPy, and imageio-ffmpeg; Python 3.9–3.12 is recommended. The font is bundled with the skill; FFmpeg is provided by imageio-ffmpeg. No need to install the original project service or execution framework. -
Read the input specification and write a
promo.jsonfile in the current working directory. Defaults:aspect=9:16,quality=preview(720×1280 / 30fps); if the user requests HD or a final video, selectfinal(1080×1920 / 60fps). Background music is enabled by default. Four looks without a brand image yields 15 seconds; with a brand image, 18 seconds; other counts are calculated automatically. -
First, generate keyframes:
PYTHON SKILL_DIR/scripts/render_video.py --spec /absolute/path/promo.json --output /absolute/path/video-output --keyframesActually inspect the returned
keyframesandtransitions, checking for look coverage, subject proportions, text readability, continuous transition frames, and brand transition. The default style is warm gray white canvas, dark capsule shapes, Song Ti titles, continuous expansion of a single rounded container, LOOK switching, collection layout, and consulting wrap-up; if a brand image is provided, it softly expands into a dark brand page. If everything looks good, proceed without asking the user for additional approval to render. -
Run the full render with the same specification and output directory:
PYTHON SKILL_DIR/scripts/render_video.py --spec /absolute/path/promo.json --output /absolute/path/video-outputThe runner copies the current assets and source code, saves
context.json, the input specification, and manifest; performs the actual render; and verifies dimensions, frame rate, duration, decoded frame count, and the required audio track. On success, returnsstage=completedwith the absolute file path. No need forframe_write_manifest,frame_finish_video, or any original project services. On failure, fix the specification or environment based on the output log; if the same failure persists twice consecutively without resolution, preserve the outputs and report the specific blocker.
Delivery and Reuse
On success, deliver the resulting MP4 along with its dimensions, duration, and frame rate; optionally include a poster, keyframes, or verification.json. Generating only keyframes does not constitute a completed video. The existence of music that can be decoded does not mean it has been listened to; do not claim audio quality has been reviewed without listening. The final video uses original photos and procedural animation; it is not a video of real people or generated by a neural network.
The output directory contains src/render_video.py, src/clothing_morph.py, src/request.json, assets, and fonts. After moving the entire output directory to another machine, install the same src/requirements.txt and run python src/render_video.py --spec src/request.json --output /new/output/path to reproduce. To change looks, edit the specification without rewriting the base animation; only design a new effect when the user explicitly requests a change.