Media Prep Workbench

Media Prep Workbench

Project home

Canonical package lives in the media-prep repo at skill/media-prep-workbench/. Prefer running scripts from that tree (or the installable media-prep CLI). Personal skill copies under ~/.codex/skills or ~/.cursor/skills should sync from the repo, not diverge.

<skill-root> below means this skill directory (repo path: <repo>/skill/media-prep-workbench).

Scope

Use this skill as the execution layer for local storefront media preparation. It is project-neutral: keep client names, product taxonomy, SEO terms, output folders, and naming decisions in the active project or task notes, not in this skill.

Default to local-only file work. Do not upload to Shopify, change product media, edit theme files, delete originals, or overwrite source folders unless the user explicitly scopes that work and the active Shopify skill or project rules allow it.

Target profile: process-images / process-videos default to --profile shopify (platform QA warnings). Use --profile generic for non-Shopify storefronts when you want the same encode path without Shopify-specific warn copy.

Output Root

When the user does not specify an output path, keep media-prep outputs under the original source directory by creating a new isolated folder:

<source-folder>\media-prep-workbench\<task-slug>-<timestamp>\

For mixed media or multiple source folders, create the run folder under the specific source folder being processed, or under the nearest common source parent when that is clearer. Use subfolders by operation so outputs do not mix with originals:

inspect-images\
inspect-videos\
processed-images\
processed-videos\
clips\
reports\
maps\

Keep source files read-only by default. Do not write processed media directly beside originals, and exclude the media-prep-workbench\ run folder from future source scans unless the user explicitly wants to reprocess previous outputs.

Workflow

1. Confirm source folder, target use, and whether the task is image, video, or mixed media.
2. Inspect before processing:
   • Run scripts/media_prep.py inspect-images for images.
   • Run scripts/media_prep.py inspect-videos for videos.
   • Review contact sheets before final naming when content-based names are needed.
3. Create a CSV name map when filenames need content-aware English names. Use columns index,output_stem (source order) and/or source_file,output_stem (exact filename); omit extensions in output_stem. Prefer the source_file key when the source folder may change between inspect and process, since index-based rows shift if files are added or removed.
4. Process into a new output folder. Preserve originals by default.
5. Verify counts, skipped duplicates, failed rows, total size reduction, contact sheets, and a CSV or JSON summary.
6. Record project-specific decisions in the project log when the surrounding project requires it.

Script Quick Start

On Windows, path summaries print as UTF-8 (the script sets the console code page when possible). If a terminal still shows mojibake for Chinese or other non-ASCII folder names, run $env:PYTHONIOENCODING='utf-8' in PowerShell before invoking the script, or use PowerShell 7+.

Image inspection:

python <skill-root>\scripts\media_prep.py inspect-images --src "E:\path\images" --out "E:\path\images\media-prep-workbench\<task-slug>-<timestamp>\inspect-images"

Image compression or enhancement:

python <skill-root>\scripts\media_prep.py process-images --src "E:\path\images" --out "E:\path\images\media-prep-workbench\<task-slug>-<timestamp>\processed-images" --name-map names.csv --enhance auto

Color-sensitive fabric texture cleanup:

python <skill-root>\scripts\media_prep.py process-images --src "E:\path\images" --out "E:\path\images\media-prep-workbench\<task-slug>-<timestamp>\processed-images" --enhance texture-safe

Shopify-consistent square images (pad to 1:1 on white, keep color faithful):

python <skill-root>\scripts\media_prep.py process-images --src "E:\path\images" --out "E:\path\images\media-prep-workbench\<task-slug>-<timestamp>\processed-images" --aspect 1:1 --fit pad --pad-color white

Rotate landscape photos 90° to portrait (4:3 landscape becomes 3:4; portrait unchanged):

python <skill-root>\scripts\media_prep.py process-images --src "E:\path\images" --out "E:\path\images\media-prep-workbench\<task-slug>-<timestamp>\processed-images" --rotate-landscape cw --enhance none

Cover-crop to 3:4 with a left-weighted focal point (keeps off-center products in frame):

python <skill-root>\scripts\media_prep.py process-images --src "E:\path\images" --out "E:\path\images\media-prep-workbench\<task-slug>-<timestamp>\processed-images" --aspect 3:4 --fit cover --focus-x 0.25 --enhance none

Per-image focal points via CSV (source_file,focus_x,focus_y; values 0..1, default center is 0.5):

python <skill-root>\scripts\media_prep.py process-images --src "E:\path\images" --out "E:\path\images\media-prep-workbench\<task-slug>-<timestamp>\processed-images" --aspect 3:4 --fit cover --focus-map maps\focus-map.csv --enhance none

After any --fit cover batch, review image-process-contact-sheet.jpg before upload. Collages, ultra-wide panoramas, and multi-subject frames often need per-image focal tuning or --fit pad instead of a single global focal point.

Video inspection:

python <skill-root>\scripts\media_prep.py inspect-videos --src "E:\path\video" --out "E:\path\video\media-prep-workbench\<task-slug>-<timestamp>\inspect-videos"

Video compression and rename (auto HDR->Rec.709 SDR for Shopify):

python <skill-root>\scripts\media_prep.py process-videos --src "E:\path\video" --out "E:\path\video\media-prep-workbench\<task-slug>-<timestamp>\processed-videos" --name-map names.csv

Optional: pad or center-crop video to a Shopify-recommended aspect ratio (16:9, 9:16, 4:3, 3:4, 1:1):

python <skill-root>\scripts\media_prep.py process-videos --src "E:\path\video" --out "E:\path\video\media-prep-workbench\<task-slug>-<timestamp>\processed-videos" --aspect 3:4 --fit cover

--fit pad (default) letterboxes; --fit cover center-crops to the target ratio with no bars.

Basic clipping:

python <skill-root>\scripts\media_prep.py trim-video --input "E:\path\video\source.mp4" --output "E:\path\video\media-prep-workbench\<task-slug>-<timestamp>\clips\clip.mp4" --start 00:00:03 --duration 8

Naming Rules

Use lowercase English slugs for upload-ready files. Prefer buyer-visible material, pattern, color, use, and shot type:

pastel-floral-jacquard-fabric-rolls-closeup

Avoid hash-only names, raw WeChat names, client-only shorthand, and unverified product claims. Use project-local terms for brand-specific collection names, SKUs, material claims, or SEO priorities.

Defaults

Use references/storefront-media-presets.md for default dimensions, video settings, enhancement presets, QA checks, verified Shopify media limits, and when to override them.

Enhancement is opt-in: process-images defaults to --enhance none (color-faithful) and --format jpg. Only enable --enhance auto|dark|medium|pastel when the user accepts tone/color adjustment, and keep none for color-sensitive products (fabric, textile, cosmetics, food, art) unless a single-image trial is approved. For fabric or textile photos where the user wants weave/pile texture clearer without shifting the colorway, use --enhance texture-safe: it adjusts only luminance contrast and unsharp masking while preserving chroma channels.

Shopify compatibility is built into the defaults so uploads are not re-processed with surprises:

• Images output at 2048 px long edge (Shopify's square recommendation); --aspect 1:1|4:5|... pads to a consistent, catalog-friendly ratio (--fit pad on white by default, --fit cover to crop). With --fit cover, use --focus-x / --focus-y (0..1, default 0.5 center) or a --focus-map CSV keyed by source_file to keep off-center products in frame. Review contact sheets after cover crops, especially for collages and ultra-wide images. A warning fires when an output drops below 1600 px because Shopify zoom degrades.
• Videos output as H.264 / yuv420p / +faststart MP4 with even dimensions, and are auto-converted from HDR (HLG/PQ) or BT.2020 to Rec.709 SDR unless --keep-hdr is set. Shopify still re-transcodes every uploaded video into its own renditions; a clean Rec.709 master only makes that transcode fast, color-accurate, and failure-free. Use --aspect to pad to a Shopify-recommended ratio (16:9, 9:16, 4:3, 3:4, 1:1).

Safety

• Preserve source files unless the user explicitly asks to overwrite or delete.
• Write outputs to a new media-prep-workbench\ run folder inside the source directory by default.
• Detect exact duplicates by SHA-256 and write skipped duplicate reports.
• Generate contact sheets and maps so a human can audit content and filenames.
• Strip metadata from generated media unless preservation is requested.
• Remove audio from product videos by default for storefront media; preserve it only when user explicitly needs sound.
• Treat AI image generation, background replacement, retouching that changes product appearance, or color changes beyond correction as separate creative work requiring explicit approval.
• Do not batch RGB autocontrast presets on pink, white, pastel, fabric, cosmetic, food, or other color-critical media without a proof image; prefer none or texture-safe.