// Remove backgrounds from images using local AI (rembg). Use when removing backgrounds from character art, mascot images, photos, or any image that needs a transparent background.
| name | bg-remove |
| description | Remove backgrounds from images using local AI (rembg). Use when removing backgrounds from character art, mascot images, photos, or any image that needs a transparent background. |
| user_invocable | true |
Remove backgrounds from images using rembg (local, offline, no data sent externally). Outputs RGBA PNG with proper transparency.
Arguments after /bg-remove:
--trim (optional) — auto-trim transparent padding after removal--output <path> (optional) — custom output path. Default: same directory, <name>-transparent.pngExamples:
/bg-remove assets/character/mascot.png/bg-remove image.png --trim/bg-remove image.png --output ~/Desktop/result.pngrembg is installed in a dedicated venv. Always activate it before use:
source ~/.claude/tools/rembg-env/bin/activate
If the venv doesn't exist, install it:
python3 -m venv ~/.claude/tools/rembg-env && source ~/.claude/tools/rembg-env/bin/activate && pip install "rembg[cpu,cli]"
Model files are cached in ~/.u2net/ (downloaded on first use per model, ~170MB for birefnet-general).
sips -g pixelWidth -g pixelHeight <path>Use the birefnet-general model — it produces the best results for character art and general images.
source ~/.claude/tools/rembg-env/bin/activate && rembg i -m birefnet-general <input> <output>
Model choice: Always use birefnet-general. It gives clean edges and handles both character art and photographic subjects well.
The Read tool renders transparency as black, so you MUST verify by compositing on a colored background:
source ~/.claude/tools/rembg-env/bin/activate && python3 -c "
from PIL import Image
import numpy as np
# Load result
img = Image.open('<output>').convert('RGBA')
alpha = np.array(img)[:,:,3]
total = alpha.size
transparent = np.sum(alpha == 0)
opaque = np.sum(alpha == 255)
print(f'Dimensions: {img.size}')
print(f'Transparent: {transparent/total*100:.1f}%')
print(f'Opaque: {opaque/total*100:.1f}%')
print(f'Corners alpha: TL={alpha[0,0]} TR={alpha[0,-1]} BL={alpha[-1,0]} BR={alpha[-1,-1]}')
# Composite on magenta for visual verification
bg = Image.new('RGBA', img.size, (255, 0, 255, 255))
bg.paste(img, (0, 0), img)
bg.save('<output_dir>/verify-magenta.png')
print('Verification image saved')
"
Then view the magenta verification image with the Read tool. The magenta should only show where background was removed.
If --trim was requested, trim transparent padding:
source ~/.claude/tools/rembg-env/bin/activate && python3 -c "
from PIL import Image
import numpy as np
img = Image.open('<output>').convert('RGBA')
alpha = np.array(img)[:,:,3]
# Find bounding box of non-transparent pixels
rows = np.any(alpha > 0, axis=1)
cols = np.any(alpha > 0, axis=0)
rmin, rmax = np.where(rows)[0][[0, -1]]
cmin, cmax = np.where(cols)[0][[0, -1]]
# Add small padding (2% of dimensions)
pad_h = max(int(img.height * 0.02), 4)
pad_w = max(int(img.width * 0.02), 4)
rmin = max(0, rmin - pad_h)
rmax = min(img.height - 1, rmax + pad_h)
cmin = max(0, cmin - pad_w)
cmax = min(img.width - 1, cmax + pad_w)
cropped = img.crop((cmin, rmin, cmax + 1, rmax + 1))
cropped.save('<output>')
print(f'Trimmed: {img.size} -> {cropped.size}')
"
Done: background removed
Source: <input_path>
Output: <output_path>
Dimensions: <width>x<height>
Transparent pixels: <percent>%
Model: birefnet-general (local, offline)
birefnet-general — best general-purpose model for this task.Test and deploy changes safely. Runs tests as a pre-deploy gate, deploys, then runs post-deploy verification. This is a TEMPLATE — customize the commands and checks for your specific deployment pipeline.
Get a second opinion from Google's Gemini Pro via the locally installed Gemini CLI (defaults to gemini-3.1-pro-preview; override with the CLAUDE_SECOND_OPINION_MODEL env var). Use this skill when in Plan Mode for large or critical tasks, when stuck on a debugging dead end, when facing architecture trade-offs, for subtle edge cases in code review, or any situation where an independent perspective would add value. Also use when the user explicitly asks for a "second opinion", "ask Gemini", "another perspective", or "cross-check this". Reports unavailability rather than falling back to a weaker model.
Update project documentation after code changes. Maintains the 4 core ai-context files (spec, project-structure, progress, deployment-infrastructure) and CLAUDE.md. Use after completing features, refactors, or any changes that affect project structure, capabilities, or status. Also creates initial documentation if files don't exist yet.
Review uncommitted code changes using parallel Claude sub-agents with specialized roles (Bug Hunter, Rules Auditor). Spawns multiple focused reviewers for large diffs (50+ lines), single reviewer for smaller changes. Checks for bugs, security issues, CLAUDE.md compliance, and test coverage gaps. Use after completing substantial implementation work, or when the Stop hook requests it. Also invocable manually with /review-work.
Edit images with precision — crop, resize, mirror, rotate, trim, and reframe. Use this skill whenever the user asks to crop, resize, trim, mirror, flip, rotate, reframe, or otherwise manipulate an image. Also use for creating square crops, portraits/headshots from full-body images, icon sizes, or any image transformation. Even if the request sounds simple, this skill prevents common pitfalls and ensures correct results on the first try.
Generate character art and image variations using AI image generation (Google Gemini) with reference images for style and character consistency. Use this skill when the user asks to generate new character poses, mascot variations, art assets, illustrations, or any AI-generated images — especially when maintaining consistency with an existing character or style.