Cloudinary Transformation Rules

When to Use

• Building Cloudinary delivery/transformation URLs
• Converting natural language requests to transformation syntax
• Debugging transformation URLs that aren't working
• Optimizing images or videos with Cloudinary
• Applying effects, overlays, resizing, or cropping

Quick Start

Default Best Practice: Always Optimize

Add fauto/qauto to the end of nearly every transformation URL (as final components):

• Automatically delivers optimal format
• Optimizes quality for best balance of visual quality and file size
• Reduces bandwidth and improves performance

Example: cfill,gauto,w400,h300/fauto/qauto

Exceptions - Don't add optimization when:

• Account has "Optimize By Default" enabled (already applied automatically)
• Special quality requirements (use qauto:best, qauto:low, or manual q_N instead)
• Specific format required (replace fauto with fpng, f_jpg, etc.)
• Delivering exact original with no modifications

Examples of common transformations (with optimization):

1. Resize: cscale,w400/fauto/qauto
2. Smart crop: cfill,gauto,h300,w400/fauto/qauto
3. Background removal: ebackgroundremoval/fpng/qauto
4. Text overlay: coyellow,ltext:Arial40:Hello%20World/fllayerapply,gsouth/fauto/qauto
5. Image overlay: llogo/cscale,w100/fllayerapply,gnorthwest,x10,y10/fauto/q_auto

Important: All transformation strings shown throughout this skill are illustrative examples to demonstrate syntax and concepts. When generating transformations, choose specific values (dimensions, colors, positions, etc.) based on the user's actual requirements and use case, not the example values shown.

For debugging: See [references/debugging.md](references/debugging.md) for detailed troubleshooting steps.

Gathering Requirements

Before generating a transformation URL, if not already specified, clarify these details based on the user's request:

For Resize/Crop Requests

Required:

• At least one dimension (width OR height)
• Crop behavior if both dimensions specified (fill, pad, scale, limit, etc.)

Clarify:

• Focal point/gravity (especially for cropping): Face detection? Center? Smart auto-detection?
• Maintain aspect ratio? (if only one dimension, this is automatic)

Example questions:

• "What dimensions do you need? (width and/or height)"
• "Should this fill the space (may crop) or fit within it (no cropping)?"
• "Any important focal point? (faces, center, specific area)"

For AI Transformation Requests

Background removal:

• Output format needs (PNG for transparency vs JPG with solid background)
• What to do with transparent area (keep transparent, add color, or gen_fill)

Generative fill:

• Target dimensions or aspect ratio
• How much extension needed

Generative replace:

• What object to replace (from)
• What to replace it with (to)
• Preserve original shape? (for clothing/objects)

Generative remove:

• What object(s) to remove
• Remove all instances or just one?

Generative background replace:

• Describe desired background (or use auto-generation)
• Need reproducibility? (consider seed parameter)

For Video Transformation Requests

Trimming:

• Start and end time, or duration
• Seconds or percentage of video

Codec/format:

• Output format needs (MP4, WebM, etc.)
• Quality requirements (use vc_auto if unsure)

Audio:

• Keep or remove audio track
• If for autoplay, suggest removing audio (ac_none)

Always Recommend

Unless user specifies otherwise:

• Add fauto/qauto at the end of transformation URLs (see Quick Start section for exceptions)
• Use g_auto for smart cropping when filling dimensions
• Consider cost for AI transformations (inform user of transformation credits)

Quick Reference

URL Structure

https://res.cloudinary.com/<cloud_name>/<asset_type>/<delivery_type>/<transformations>/<version>/<public_id>.<ext>

Key Rules:

• Commas (,) separate parameters within a component
• Slashes (/) separate components between transformations
• Each component acts on the output of the previous one

Parameter Types

Action parameters: Perform transformations (one action per component: each action transformation should be separated by a slash) Qualifier parameters: Modify action behavior (in the same component as the action, using commas as separators)

Check the Transformation Reference to determine if a parameter is an action or qualifier.

Core Transformations

Resize & Crop

Dimension value formats:

• Whole numbers (e.g., w400, h300) = pixels
• Decimal values (e.g., w0.5, h1.0) = percentage of original dimensions (0.5 = 50%, 1.0 = 100%)

Choosing the right crop mode:

Use c_scale when:

• Resizing while maintaining original aspect ratio
• Specify only ONE dimension (width OR height)
• No cropping needed
• The user intentionally wants to stretch or squash an image by changing the aspect ratio

Use c_fill when:

• Must fit exact dimensions (e.g., thumbnail grid, fixed layout)
• Okay to crop parts of image
• Combine with gauto for smart cropping, or gface for portraits

Use c_fit when:

• Image must fit within dimensions without cropping
• Okay to have empty space
• Maintaining full image content is critical

Use c_pad when:

• Must fit exact dimensions without cropping
• Need to fill empty space with background color/blur (videos only)/AI-generated pixels
• Use with b<color>, bauto, bblurred (blurred background - videos only), or bgen_fill

Use c_limit when:

• Set maximum dimensions but don't upscale small images
• Preserving original quality of small images matters

Use c_thumb when:

• Creating thumbnails (typically avatars)
• Use with g_face for face-centered crops

Use c_auto when:

• Cloudinary should intelligently crop to interesting content
• Combine with g_auto for best results
• Good for dynamic content where focal point varies

Examples:

c_scale,w_400                      # Resize width to 400px, maintain aspect ratio
c_scale,w_0.5                      # Resize to 50% of original width
c_fill,g_auto,h_300,w_400          # Fill 400x300px dimensions, smart crop
c_fit,h_300,w_400                  # Fit within dimensions, no crop
c_pad,b_white,h_300,w_400          # Pad to exact size with white background
c_pad,w_1.0                        # Pad to original width (100%)
c_limit,w_1000                     # Limit max width, no upscale
c_thumb,g_face,h_150,w_150         # Face-centered square thumbnail
c_auto,g_auto,w_800                # Auto crop to interesting area

Important: Always specify a crop mode explicitly. Avoid using both dimensions with c_scale (will distort if aspect ratios don't match) - prefer one dimension to maintain aspect ratio.

Gravity (Focal Point)

Gravity determines which part of the image to focus on when cropping:

• g_auto - Smart detection (recommended for varied content; detects faces, objects, contrast)
• g_face - Face detection (portraits, avatars)
• g_center - Center position (centered subjects, logos)
• gnorth, gsouth_east, etc. - Compass positions (fixed locations, overlay positioning)
• xN,yN - Custom offsets (integers = pixels, floats = percentage: 0.8 = 80%)

Examples:

c_fill,g_auto,w_400,h_300                      # Smart crop
c_thumb,g_face,w_200,h_200                     # Face-centered
l_logo/fl_layer_apply,g_south_east,x_10,y_10  # Logo bottom-right

Important:

• gauto only works with cfill, clfill, ccrop, cthumb, cauto
• When using x, y, h, w together, use all integers OR all floats (don't mix)

Format & Quality

Recommended defaults:

• fauto/qauto - Use for most production images (WebP to supported browsers, optimized file size)

Specific formats (when requirements dictate):

• f_png - Transparency needed (e.g., after background removal)
• f_jpg - Force JPEG (remove transparency)
• qN - Manual quality 1-100 (e.g., q60 for thumbnails, q_90 for hero images)
• dpr_auto - Retina displays (Chromium-only, requires Client Hints - see limitations below)

Examples:

f_auto/q_auto           # Recommended default
f_png/q_auto            # PNG with transparency
q_80                    # Manual 80% quality

Best Practice: Use / to separate format and quality as distinct components.

Responsive Images (dprauto, wauto)

dpr_auto - Automatically adapts to device pixel ratio (Retina displays)

• Chromium-only (Chrome, Edge, Opera, Samsung Internet)
• Requires Client Hints configuration
• Falls back to dpr_1.0 on other browsers
• Does NOT work inside named transformations

Alternative for universal support: Use explicit dpr_2.0 or <img srcset> with 1x/2x variants

For Client Hints configuration, browser compatibility, responsive breakpoints, and framework integration, see [references/responsive-images.md](references/responsive-images.md)

Effects

Common effects:

• e_grayscale - Black and white (artistic, accessibility)
• e_sepia - Vintage/nostalgic feel
• e_blur:N - Blur (privacy, placeholders; N typically 300-2000)
• e_sharpen - Enhance clarity (useful after resizing)
• e_cartoonify - Illustrated style
• corgb:RRGGBB,ecolorize:N - Color tint (N = intensity 0-100, for brand theming)
• ebackgroundremoval - See AI Transformations section

Examples:

e_blur:800                       # Blur effect
e_sharpen                        # Enhance clarity
co_rgb:0044ff,e_colorize:40      # Blue tint at 40%

Note: Color (co) is a qualifier - use in same component as ecolorize.

Overlays & Underlays

Use for:

• l<publicid> - Image overlays (logos, watermarks, badges)
• u<publicid> - Image underlays (custom backgrounds behind transparent subjects)
• ltext:fontsize:text - Text overlays (labels, social cards, dynamic text)

Pattern:

1. Declare: l<publicid> or u<publicid> or ltext:Arial40:Hello%20World
2. Transform (optional): e.g. /cscale,w100/ or /o_50/ (opacity)
3. Apply: /fllayerapply,g<position>,x<offset>,y_<offset>

Critical: Using fl_relative for overlay dimensions:

• Without fl_relative: Dimensions are relative to the overlay's original size

- Example: w_1.0 = 100% of the overlay image's width (not useful for small images)

• With fl_relative: Dimensions are relative to the base image's size

- Example: w1.0 = 100% of the base image's width (covers entire width) - Always use flrelative when sizing overlays as a percentage of the base image

Examples:

l_logo/c_scale,w_100/fl_layer_apply,g_north_west,x_10,y_10                # Logo at 100px
l_logo/c_scale,fl_relative,w_0.25/fl_layer_apply,g_north_west,x_10,y_10  # Logo at 25% of image width
l_docs:one_black_pixel/c_scale,fl_relative,h_1.0,w_1.0/o_50/fl_layer_apply # Full-image semi-transparent overlay
co_yellow,l_text:Arial_40:Hello%20World/fl_layer_apply,g_south            # Text overlay
u_background/e_background_removal                                          # Custom background
c_fill,h_400,w_300/l_same_image/c_fill,e_grayscale,h_400,w_300/fl_layer_apply,g_west,x_300 # Side-by-side (600×400)

Important:

• Color (co_) is a qualifier — use in the same component as text overlay declaration
• Always use fl_relative when you want overlay dimensions as a percentage of the base image
• Side-by-side / canvas extension: to place an overlay beside the base, offset it past the base edge — the canvas auto-expands. Use gwest,x<basewidth> for horizontal or gnorth,y<baseheight> for vertical.

Borders & Rounding

• r_N - Rounded corners (N = radius in pixels; for modern UI, cards)
• r_max - Perfect circle (use with square dimensions; avatars, icons)
• boNNpxsolid_color - Border (frame images, separate from background)

Examples:

r_20                           # 20px rounded corners
r_max                          # Perfect circle
bo_5px_solid_black             # 5px black border
r_20,bo_5px_solid_rgb:0066ff   # Rounded with border (same component)

Important: For borders that follow rounded corners, use border as qualifier in same component.

Background Color

• bcolor,cpad - Fill empty space with solid color (product images, letterboxing)
• bauto,cpad - Aautomatically selected background color based on one or more predominant colors in the image
• bgenfill,c_pad - AI-extended background (change aspect ratio without cropping; see AI Transformations for cost)

Examples:

b_lightblue,c_pad,w_1.0         # Light blue background
b_auto,c_pad,ar_16:9            # Automatically selected color for background, 16:9
b_gen_fill,c_pad,ar_1:1         # AI-extended to square
b_blurred,c_pad,ar_16:9         # Blurred background (videos only), 16:9

Critical: Background (b) is a qualifier - use with pad crop in same component: bcolor,cpad,wX, NOT /b_color/.

Rotation & Flips

• a90, a180, a_270 - Rotate in 90° increments (correct orientation)
• aN - Rotate by degrees (e.g., a-2 to straighten crooked photos)
• a_hflip - Horizontal flip (mirror selfies, directional images)
• a_vflip - Vertical flip (reflections)
• aautoright/aautoleft - Auto-rotate based on EXIF orientation

Examples:

a_90                    # Rotate 90° clockwise
a_-2                    # Straighten slight tilt
a_hflip                 # Mirror horizontally
a_auto_right            # Auto-fix from EXIF

Asset Type Matters (Image vs. Video)

Many flags and parameters apply to only one asset type. Applying one to the wrong base often fails silently — the URL still returns a valid 200 with no X-Cld-Error, just the wrong output. This goes both ways: video-only syntax on an image, and image-only syntax on a video. Always verify the actual output (dimensions, duration, frame count) rather than assuming it worked.

Common video-only examples (this is not an exhaustive list — ~35 parameters are video-only):

• flsplice (flag) - Concatenate a clip/image onto the video timeline (no image equivalent — to place media side-by-side, offset the overlay to extend the canvas: fllayerapply,gwest,x<basewidth>)
• du, so, eo_ - Trim/seek by time (duration, start offset, end offset)
• fps_ - Set frame rate
• vc, ac - Video / audio codec
• eboomerang, eprogressbar - Video-only effects

When unsure whether a flag or parameter supports your asset type, check the Transformation Reference before applying it.

Named Transformations

Named transformations (t_<name>) save transformation chains for reuse. Suggest for:

• Transformations used across multiple assets
• Complex transformation chains
• Expensive operations (to enable baseline transformations and reduce costs)

Baseline transformations (bl<name>) cache expensive named transformations so they don't need to be regenerated. Use bl instead of t_ for AI transformations (background removal, generative AI) that will have variations applied. This can reduce costs from 75-230 tx per variation down to 1 tx each after the initial baseline is generated.

Example: blbgremoved/cscale,w500 - Uses cached background removal result, only pays for resize (1 tx instead of 75 tx)

Important: fauto, dprauto, and wauto don't work inside named transformations - use them directly in URLs: tavatar/fauto/qauto

For complete details, limitations, and baseline transformation examples, see [references/named-transformations.md](references/named-transformations.md)

Generative AI Transformations

Proactively suggest these AI transformations when appropriate:

Note: Numbers in parentheses (e.g., 75 tx) indicate additional transformation credits consumed per use. Standard transformations = 1 tx.

• ebackgroundremoval (75 tx) - Remove backgrounds (e-commerce, profiles; combine with fpng or bcolor,c_pad)
• bgenfill (50 tx) - Extend backgrounds (change aspect ratio without cropping; use with c_pad)
• egenbackgroundreplace:prompt<text> (230 tx) - AI-generated backgrounds (custom environments, seasonal variations; high cost)
• egenreplace:from<obj>;to<new> (120 tx) - Swap objects (product variations, colors; use ;preservegeometrytrue for clothing)
• egenremove:prompt_<text> (50 tx) - Remove objects (clean up distractions)
• eautoenhance (100 tx) - Improve quality (fix poor lighting/exposure)
• e_upscale (10-100 tx) - Enlarge without quality loss (low-res to high-res)

Important: AI transformations cost significantly more (50-230 tx vs 1 tx). Inform users of costs and consider baseline transformations (e.g., blbgremoved/cscale,w500) to avoid re-processing expensive operations - see [references/named-transformations.md](references/named-transformations.md#baseline-transformations) and [references/transformation-costs.md](references/transformation-costs.md) for details.

For complete details, syntax, and powerful combinations, see [references/ai-transformations.md](references/ai-transformations.md)

Video-Specific Transformations

Critical: Use fauto:video (not just fauto) to ensure video output - plain f_auto may return an image thumbnail.

• vc_auto - Automatic codec (recommended; optimal for browser/device)
• soN/eoM - Trim (start/end in seconds; create clips, remove intro/outro)
• ac_none - Remove audio (essential for autoplay; reduces file size)
• fps_N - Set frame rate (lower = smaller file; standardize rates)
• Video resizing - Same crop modes as images (cfill, cscale, c_pad)

Common patterns:

vc_auto/ac_none/f_auto:video/q_auto                      # Autoplay-ready
so_0/du_10/vc_auto/f_auto:video/q_auto                   # First 10 seconds
c_scale,w_720/vc_auto/f_auto:video/q_auto                # Resize to 720p width
c_fill,g_auto,h_720,w_1280/vc_auto/f_auto:video/q_auto  # 720p HD, smart crop

For complete details including codecs, trimming strategies, and video concatenation, see [references/video-transformations.md](references/video-transformations.md)

Variables & Conditionals

Variables reuse values and create templates:

$size_300/c_fill,h_$size,w_$size        # Reuse value
$iw/w_$iw_div_2                         # Half original width (arithmetic)

Conditionals adapt transformations dynamically:

if_w_gt_1000/c_scale,w_1000/if_end                          # Responsive sizing
if_ar_gt_1.0/c_fill,w_800,h_450/if_else/c_fill,w_450,h_800/if_end  # Orientation handling

Key rules:

• Variable names: alphanumeric, start with letter, no underscores
• Conditionals: Must close with if_end
• Arithmetic: add, sub, mul, div (left-to-right evaluation)

For complete syntax, arithmetic operations, nested conditionals, and real-world patterns, see [references/advanced-features.md](references/advanced-features.md)

Self-Validation Checklist

Before returning a transformation URL, verify:

1. ✅ URL structure is complete (cloudname, assettype /image/ or /video/ or /raw/, deliverytype, publicid)
2. ✅ Each component has only one action parameter (e.g., one crop mode per component)
3. ✅ Crop mode is explicit (don't rely on defaults; avoid both dimensions with c_scale)
4. ✅ Overlays end with fllayerapply in separate component
5. ✅ Text strings are URL-encoded (spaces = %20, special chars encoded)
6. ✅ Variable names follow rules (alphanumeric, start with letter, no underscores)
7. ✅ gauto compatibility (only works with cfill, clfill, ccrop, cthumb, cauto)
8. ✅ Background as qualifier (use with pad crop: bcolor,cpad,wX, not /bcolor/)
9. ✅ Format/quality at end (prefer fauto/qauto as final components)
10. ✅ Flags/parameters match the base asset type (asset-type-specific syntax — e.g. video-only flsplice, du, fps, vc — often no-ops silently on the wrong base, in either direction; verify the output and check the Asset Type Matters section above)
11. ✅ Transformation parameters are valid (don't make up any parameter names - check against Transformation Reference)

Quick syntax check:

• Commas separate parameters within a component: cfill,gauto,w_400
• Slashes separate components: cfill,w400/fauto/qauto
• Actions vs qualifiers: Only one action per component, qualifiers modify that action

See [references/debugging.md](references/debugging.md) for detailed examples of each check.

Debugging Checklist

When a transformation isn't working:

1. Verify URL structure: Check that all required URL parts are present:

- Cloud name: /<cloud_name>/ - Asset type: /image/ or /video/ or /raw/ - Delivery type: /upload/ or /fetch/ etc. - Public ID at the end

1. Check the X-Cld-Error header: Cloudinary reports errors in the X-Cld-Error HTTP response header
2. Check parameter names against Transformation Reference
3. Check crop mode: Specify crop mode explicitly; avoid both dimensions with c_scale (causes distortion if aspect ratios don't match)
4. Verify gravity compatibility: gauto doesn't work with cscale, cfit, climit, c_pad
5. Check action vs qualifier: Only one action per component, qualifiers in same component
6. Verify overlay pattern: Must end with fllayerapply component
7. Check variable names: No underscores, must start with letter
8. Verify URL encoding: Text overlays need URL-encoded strings (spaces = %20)
9. Check auto parameters in named transformations: fauto, dprauto, and w_auto don't work inside named transformations - use them directly in URLs
10. Verify Client Hints for dprauto/wauto: These only work on Chromium browsers with Client Hints enabled; fallback to dpr_1.0 otherwise (see [references/responsive-images.md](references/responsive-images.md) for configuration)
11. Video returns image instead of video: Use fauto:video (not just fauto) for video transformations - plain f_auto may return an image thumbnail

Checking X-Cld-Error Header

The X-Cld-Error header contains error details when a transformation fails. To check it:

Using browser DevTools:

1. Open Developer Tools (Network tab)
2. Request the transformation URL
3. Look for X-Cld-Error in Response Headers

Using code (fetch the URL):

fetch('https://res.cloudinary.com/demo/image/upload/w_abc/sample.jpg')
  .then(response => {
    const error = response.headers.get('x-cld-error');
    if (error) {
      console.log('Cloudinary Error:', error);
    }
  });

Common X-Cld-Error messages:

• Invalid width - abc - Width parameter expects a number
• Invalid transformation syntax - Malformed transformation string
• Resource not found - Asset doesn't exist or public ID is incorrect
• Transformation limit exceeded - Account transformation quota reached

Online tool: Use the X-Cld-Error Inspector to check any Cloudinary URL

For more details, see Error Handling

Transformation Costs

Important: Warn users about high-cost transformations before generating URLs. AI effects cost significantly more than standard transformations (50-230 tx vs 1 tx).

For complete cost details and cost reduction strategies, see [references/transformation-costs.md](references/transformation-costs.md)

Additional Resources

Skill References (Progressive Disclosure)

• [references/debugging.md](references/debugging.md) - Use when transformations return errors or unexpected results
• [references/ai-transformations.md](references/ai-transformations.md) - Use when you need AI transformation prompt syntax, cost details, or complex AI combinations
• [references/video-transformations.md](references/video-transformations.md) - Use when working with video codecs, trimming strategies, concatenation, or creating animated images from videos
• [references/advanced-features.md](references/advanced-features.md) - Use when building complex logic with variables, conditionals, or arithmetic
• [references/responsive-images.md](references/responsive-images.md) - Use when implementing responsive images, configuring Client Hints, or using dprauto/wauto
• [references/transformation-costs.md](references/transformation-costs.md) - Use when optimizing for cost or explaining cost implications to users
• [references/named-transformations.md](references/named-transformations.md) - Use when creating reusable transformations or reducing costs for repeated operations
• [references/examples.md](references/examples.md) - Use when you need real-world examples beyond the Quick Start (social cards, e-commerce, responsive images)

Core Cloudinary Documentation

• Transformation Reference - All parameters

Image Transformations

• Image Transformations Overview
• Resizing and Cropping
• Placing Layers on Images
• Effects and Enhancements
• Background Removal
• Generative AI Transformations
• Face-Detection Based Transformations
• Custom Focus Areas
• Transformation Refiners
• Animated Images
• Transformations on 3D Models
• Conditional Transformations
• User-Defined Variables and Arithmetic
• Custom Functions

Video Transformations

• Video Transformations Overview
• Resizing and Cropping
• Trimming and Concatenating
• Placing Layers on Videos
• Effects and Enhancements
• Audio Transformations
• Converting Videos to Animated Images
• Conditional Transformations
• User-Defined Variables and Arithmetic

Common Mistakes & Best Practices

Avoid:

• ❌ w400,h300 → ✅ cscale,w400 (both dimensions with c_scale distorts image; prefer one dimension)
• ❌ cscale,gauto,w400 → ✅ cfill,gauto,w400 (gauto doesn't work with cscale)
• ❌ llogo/fllayerapply,gnorthwest → ✅ llogo/cscale,w100/fllayerapply,gnorthwest
• ❌ blightblue/etrim → ✅ blightblue,cpad,w1.0/etrim (background as qualifier)

Always:

• Prefer fauto/qauto in separate components over fauto,qauto
• Use g_auto for smart cropping unless specific focal point needed
• Specify crop mode with width/height; prefer one dimension with c_scale
• Never guess parameter names - verify against documentation