Section Dividers

Generate transparent decorative dividers for web page sections.

CRITICAL RULES - Read Before Generating

1. FULL WIDTH Required

• Strip MUST extend from LEFT EDGE to RIGHT EDGE of image
• NO floating islands - content spans entire width
• Add "FULL WIDTH from left edge to right edge" to prompt

2. Cross-Section NOT Reflection

• TOP edge: Content pointing UPWARD (houses, trees, rooftops)
• BOTTOM edge: DIFFERENT content pointing DOWNWARD (pipes, caves, roots)
• NOT a mirror/reflection of the same scene
• Add "TOP EDGE has [X] pointing UPWARD. BOTTOM EDGE has [Y] pointing DOWNWARD" - use DIFFERENT X and Y

3. SOLID Filled Strip

• Strip must be SOLID between top and bottom edges
• NOT hollow, NOT a cave opening, NOT an arch
• Add "SOLID FILLED strip - not hollow, not a cave"

4. Clean Boundary Required

• Sharp, clear edge between silhouette content and background
• High contrast between the two colors
• If edges are fuzzy or blended = generation failed

5. Content Must NOT Touch Image Edges

• Content should float in CENTER with background above and below
• If content touches TOP or BOTTOM of image = JUNK, do not process
• Verify raw image BEFORE running bg removal

6. Exact Color Matching

• Background = EXACT parallax color (use analyze-bg.ts)
• Silhouette = EXACT section CSS color
• Both colors must be explicitly stated in prompt with hex values

Background Color Strategy

Use a HIGH CONTRAST color from the parallax image's own palette. This ensures:

1. Clean edges when removing background (high contrast = sharp removal)
2. No color fringing or halos (color exists in the parallax)
3. Seamless blending with the actual parallax

Get the Parallax Color

# Analyze the parallax image to find colors in its palette
bun run ${CLAUDE_PLUGIN_ROOT}/skills/section-dividers/scripts/analyze-bg.ts /path/to/parallax.png top   # for top border
bun run ${CLAUDE_PLUGIN_ROOT}/skills/section-dividers/scripts/analyze-bg.ts /path/to/parallax.png bottom # for bottom border

Choose the color from the parallax palette that provides HIGHEST CONTRAST with your silhouette color. For dark silhouettes, pick the lightest color from the parallax. For light silhouettes, pick the darkest.

Background Removal

Use remove-bg.ts (AI-based) for reliable background removal:

bun run ${CLAUDE_PLUGIN_ROOT}/skills/section-dividers/scripts/remove-bg.ts input.png output.png

Generation Workflow

Step 1: Generate

bun run ${CLAUDE_PLUGIN_ROOT}/skills/generate-image/scripts/generate.ts "[PROMPT]" --aspect 21:9 --size 2K --negative "[NEGATIVES]" --output divider-raw.png

Prompt template:

ONE THIN horizontal SOLID terrain strip floating in solid [PARALLAX_COLOR].
FULL WIDTH from left edge to right edge.
The strip is SOLID FILLED - not hollow, not a cave.
TOP EDGE silhouette: [SURFACE_CONTENT] pointing UPWARD.
BOTTOM EDGE silhouette: [UNDERGROUND_CONTENT] pointing DOWNWARD.
SOLID MASS between top and bottom edges - filled with [SECTION_COLOR].
JAGGED organic edges on BOTH top AND bottom.
Strip is 12-15% of image height floating in vertical center.
85% solid [PARALLAX_COLOR] background above and below.
Shadow puppet silhouettes.

Negative prompt:

hollow, cave, empty center, arch, tunnel view, flat edges, reflection, mirror, white, pink, magenta, tall, thick, floating island, not full width

Step 2: Verify Raw Image (CRITICAL)

Before processing, verify:

• Content spans FULL WIDTH (left to right)
• TOP content points UP, BOTTOM content points DOWN (not reflection)
• Strip is SOLID (not hollow/cave)
• Content does NOT touch top/bottom image edges
• Clear boundary between content and background

If ANY check fails = REGENERATE. Do not process junk.

Step 3: Remove Background

bun run ${CLAUDE_PLUGIN_ROOT}/skills/section-dividers/scripts/remove-bg.ts divider-raw.png divider.png

Uses cjwbw/rembg model which PRESERVES RESOLUTION.

Step 4: Verify Final Image

• Transparency is clean (no artifacts)
• Full resolution preserved
• Blends with parallax and section on page

Scripts

remove-bg.ts

AI background removal via Replicate (rembg model). Preserves original resolution.

bun run ${CLAUDE_PLUGIN_ROOT}/skills/section-dividers/scripts/remove-bg.ts input.png output.png

analyze-bg.ts

Extract dominant color from parallax image.

bun run ${CLAUDE_PLUGIN_ROOT}/skills/section-dividers/scripts/analyze-bg.ts parallax.png bottom

colorize.ts

Tint transparent silhouette to match theme.

bun run ${CLAUDE_PLUGIN_ROOT}/skills/section-dividers/scripts/colorize.ts input.png output.png "#042f2e"

Context Discipline

Do not read generated divider images back into context. Scripts output file paths. Ask the user to visually verify the divider in their browser against the parallax and section backgrounds. To inspect programmatically, optimize the image first (via the optimize-images skill).

Fixing Existing Dividers (Color Mismatch)

When an existing divider has wrong colors or semi-transparent edges:

Step 1: Remove Background

bun run ${CLAUDE_PLUGIN_ROOT}/skills/section-dividers/scripts/remove-bg.ts original.png transparent.png

Step 2: Make SOLID with Section Color

The remove-bg output has semi-transparent edges. To make it SOLID and match the section:

from PIL import Image

img = Image.open('transparent.png').convert('RGBA')
pixels = img.load()
width, height = img.size

# Get EXACT section color from CSS/computed style
# Example: oklch(0.12 0.015 260) = RGB(3, 6, 11)
sr, sg, sb = 3, 6, 11

for y in range(height):
    for x in range(width):
        r, g, b, a = pixels[x, y]
        if a > 50:  # Any visible pixel = SOLID section color
            pixels[x, y] = (sr, sg, sb, 255)
        else:  # Transparent stays transparent
            pixels[x, y] = (0, 0, 0, 0)

img.save('final.png')

Key points:

• Alpha > 50 = make fully opaque (255) with section color
• Alpha <= 50 = make fully transparent (0)
• This eliminates semi-transparent edges from AI removal
• Section color must EXACTLY match the adjacent section's background

Getting Exact Section Color

// In browser console on the page:
const el = document.querySelector('#section-id');
const canvas = document.createElement('canvas');
canvas.width = 1; canvas.height = 1;
const ctx = canvas.getContext('2d');
ctx.fillStyle = getComputedStyle(el).backgroundColor;
ctx.fillRect(0, 0, 1, 1);
const data = ctx.getImageData(0, 0, 1, 1).data;
console.log(`RGB(${data[0]}, ${data[1]}, ${data[2]})`);

Troubleshooting

CSS Positioning

Animated vs Static Dividers

For ANIMATED dividers (with .float or other CSS animations):

• Use fixed pixel values: top: -20px
• DO NOT use translateY(-50%)

For STATIC dividers (no animation):

• translateY(-50%) works fine
• Centers element on the seam

Why Transforms Fail with Animations

When an element has a CSS animation that uses transform in its keyframes:

@keyframes float {
  0%, 100% { transform: translateY(0px); }
  50% { transform: translateY(-10px); }
}

The animation's transform completely overrides any inline transform:

• Your style={{ transform: "translateY(-50%)" }} gets replaced
• Element renders with animation's translateY(0px) instead

Also note: percentage margins (margin-top: -50%) are based on parent WIDTH, not height.

Correct Approach: Simple Negative Top

/* Position divider on seam between sections */
.divider-top {
  position: absolute;
  top: -20px;  /* Adjust until CENTER sits on seam */
  left: 0;
  right: 0;
  z-index: 50;
  pointer-events: none;
}

Layout Structure

SECTION (solid background)
    ↓
    SEAM ← divider CENTER should align HERE
    ↓
PARALLAX CONTAINER
  └── DIVIDER (absolute, top: -20px, adjusts per image)
  └── PARALLAX IMAGE
  └── DIVIDER (absolute, bottom: 0, transform for bottom)
    ↓
    SEAM
    ↓
SECTION (solid background)

Implementation Pattern

From tokenpass-web/src/components/BorderedParallax.tsx:

{/* Top divider - sits on seam above parallax */}
<div
  className="absolute left-0 right-0 z-50 pointer-events-none float"
  style={{ top: "-20px" }}  // Simple fixed value
>
  <img src={borders.top} className="w-full h-auto" />
</div>

Positioning Tips

1. Start with top: -20px and adjust visually
2. Goal: CENTER of divider image sits on the SEAM
3. If image has transparent space, account for it
4. Test on both mobile AND desktop
5. Fixed pixel values work consistently across screen sizes