Smart BG Removal & Sharpening

What It Solves

Background removal is not just deleting the background. Portraits can lose hair strands, product shots can have rough edges, and character art often misses detached attachments such as swords, ribbons, or tails. A generic rembg run or API call may produce a usable PNG, but it may still need matting cleanup, upscaling, and sharpening. smart-bg-removal-sharpen targets these cases by combining standard removal, hair refinement, multi-subject extraction, and image enhancement in one local scripting workflow.

How It Works

The skill prefers the unified entrypoint scripts/run_bg_removal.py and selects a precision tier:

• Standard: solid backgrounds, logos, and products; fast
• Refined: hair, fur, and semi-transparent edges; useful for portraits and pets
• High precision: anime characters, complex costumes, and game posters; can use BRIA RMBG-2.0

When the request is to extract “the left character,” “the right character,” or keep “the headband and weapon,” it switches to a multi-subject extraction flow. It can fuse masks from birefnet-portrait, birefnet-general, and nearby attachments via --max-gap. If the asset needs to look sharper, the workflow can call upscale_image.py, sharpen_image.py, or enhance_image.py, and still output transparent PNGs. Batch jobs also generate batch_summary.md and batch_summary.json for success/failure review.

Boundaries

This is best for design media, avatars, e-commerce images, and character assets. Online APIs for Remove.bg, Clipdrop, and Tencent portrait services require configured keys and may have quota limits. The BRIA weights are CC BY-NC 4.0, so non-commercial use is the safer assumption. First runs may download models, and batch jobs on large images need enough disk space and runtime.