l-make-product-photo

Create product photos with background removal and standardized formatting. Use when: (1) User says '/l-make-product-photo' or 'make product photo', (2) User wants to create product photos, remove back...

Make Product Photo

Create standardized product photos from source images using ML-based background removal.

What It Does

• Removes the background from product images using ML-based detection
• Creates a 1600x1600 square JPEG with:
  • Default background: fabric texture image (bundled at assets/default-bg.jpg)
  • Override with --with-bg <path> for a different background
  • Product centered in a 90% container (1440x1440 area)
  • Product fitted like CSS object-fit: contain
• Outputs to __inbox/prod-imgs-exported/ by default

Usage

node ./sub-packages/product-photo-maker/bin/make-product-photo.mjs <image-path> [<image-path-2> ...] [--output <dir>]

Examples

Single image:

node ./sub-packages/product-photo-maker/bin/make-product-photo.mjs ./imgs/my-product.heic

Multiple images:

node ./sub-packages/product-photo-maker/bin/make-product-photo.mjs ./imgs/photo1.jpg ./imgs/photo2.png ./imgs/photo3.heic

Custom output directory:

node ./sub-packages/product-photo-maker/bin/make-product-photo.mjs ./imgs/photo.jpg --output ./my-output-dir

Options

When NOT to use this skill

Skip this skill entirely when the source is an already-taken final studio photo — top-down, on the Takazudo orange fabric, properly framed, panels/parts where you want them. The photo is already “taken”; the only step needed is /l-orange-tweak for tone normalization, then drop the file into /imgs/.

Synthesizing shadow on a flat top-down photo darkens the corners with a fake vignette instead of producing depth — the image looks worse, not better. The shadow stack here was built for the “incoming photo / bg-removed composite” path: take a non-orange or stock manufacturer photo, remove the background, composite onto orange fabric, then synthesize lighting + shadow. That synthesis makes a panel-cutout look like a real photo. Applying the same synthesis to a real photo is double-application.

Decision: Was this photo already taken (top-down, on the orange fabric, framed and lit)?

• Yes → skip this skill. Run /l-orange-tweak only, then /imgs/ → /l-metadata-update.
• No, it needs synthesis → continue below.

Deciding flags per image

When the user provides images, look at each image to decide which flags are needed. Since flags apply to all images in a single run, group images by mode and run the command separately for each group.

--no-bg-removal (skip background removal)

Use when the image IS the product with no background — the product fills the entire frame edge-to-edge. Background removal would damage the product colors (especially dark/black products where the ML model confuses the product with background).

--no-trim (skip trimming)

Use when background removal is needed but the product already fills most of the frame. Trimming could over-crop the product.

Default (no flags)

Use when the image has a clear, visible background (white, gray, colored, studio backdrop, etc.) surrounding the product. Background removal extracts the product, trimming removes leftover transparent space so the product fills the container.

Decision flow

1. Product fills entire frame, no background visible → --no-bg-removal
2. Product on a background, product fills most of the frame → --no-trim
3. Product on a background with significant space around it → default (no flags)

Shadow decision (--shadow and --float)

Once you’ve decided this skill should run at all (i.e., the photo needs synthesis — see “When NOT to use this skill” above), add --shadow and decide float vs grounded:

• Modular synth panels (Eurorack modules with visible panel edges, jacks, knobs on a flat faceplate) → --shadow --float (panel hovers above surface)
• Desktop/standalone products (MIDI controllers, desktop units, boxes that sit flat) → --shadow only (grounded shadow)

AI detection hint: Modular synth panels are typically tall-narrow rectangles (3

When unsure, ask the user: “Is this a modular synth panel (floating) or a desktop product (grounded)?” to decide --float.

Notes

• Accepts jpg, png, webp, heic and other common image formats
• Output filename matches input filename with .jpg extension
• First run downloads the ML model (~30MB), subsequent runs are faster
• If dependencies are not installed, run: cd sub-packages/product-photo-maker && pnpm install
• Output directory is created automatically if it does not exist

Workflow Integration

After creating product photos, you may want to:

1. Move images to /imgs/ for the image pipeline
2. Run /l-metadata-update to process and upload to R2
3. Reference the images in product master data or MDX content