
A Cloudinary layer sized with fl_relative follows the base image, not a detected face. Use fl_region_relative with an appropriate region gravity and a decimal size when the layer must track that region. Neither flag proves that the requested face or other region was correctly detected.
Key Takeaways
- Ordinary decimal sizing starts from the overlay's own source size.
fl_relativeuses the base image as the size reference.fl_region_relativeuses the region selected by the gravity type.
Decide which object defines the size
Cloudinary's layer-transformation documentation distinguishes three references. A relative resize normally refers to the overlay's original dimensions. Adding fl_relative makes decimal width or height relative to the base. Adding fl_region_relative makes it relative to a region selected with a supported gravity, such as g_face, g_faces, g_ocr_text or g_custom.
Choose the reference from the editorial requirement. A corner brand mark may need to follow the canvas; a temporary face-covering review badge may need to follow the detected face. These are different contracts even when they happen to look similar on one portrait.
Construct one controlled comparison
- Keep the clean approved source and a non-sensitive comparison copy.
- Record whether the badge should follow canvas width or a selected region.
- For canvas-relative sizing, use
fl_relativewith the intended decimal width or height. - For region-relative sizing, use
fl_region_relative, a decimal dimension and the appropriate gravity. - Close the layer with
fl_layer_applyand the intended placement. - Compare portraits with visibly different face sizes and positions.
Cloudinary's documented face-covering example uses fl_region_relative,w_1.3 and g_faces. That is an example of sizing to detected regions, not a promise that 130% covers every identity cue.
A hypothetical group-photo check
Suppose a studio needs review badges over faces of different sizes. A badge scaled to 20% of the whole canvas can look large on a distant person and small on a close face. Region-relative sizing addresses the reference mismatch, but the reviewer still needs to inspect every intended region.
Use a proposed matrix with one close portrait, one distant portrait and a multi-person frame. Record detection, badge size and visible uncovered features separately. No result from this matrix is claimed here.
Acceptance and limits
Accept only when the output places the correct badge on every intended subject at the approved size. If detection misses someone, stop automatic delivery and choose a supported deliberate region or another reviewed workflow.
The portrait crop-anchor guide concerns framing people consistently. This issue concerns a layer's mathematical size reference. Do not describe a visual badge as verified anonymization without separately reviewing the final image and the disclosure requirement.
Related tasks
- Keep a Lightroom Watermark Readable Across Different Crops
- How Do I Crop a Set of Portraits Consistently When Everyone Is a Different Height?
- Why Does My Image Edit Land in the Wrong Place When I Copy Preview Coordinates?
Sources and scope
- Cloudinary: Transforming layers: Chained operations before layer_apply target the last open layer; placement at closing; nested closing order; relative sizing to overlay, base or detected region.
Documentation-based editorial guidance. All scenarios, worksheets and acceptance checks are hypothetical or proposed. No hands-on test, account operation, benchmark, measured search demand, AI-citation rate or platform acceptance is claimed.
Prepared by Imagild Editorial, the company editorial team responsible for these guides. AI assisted research and drafting. This is documentation-based guidance with proposed examples, not a claim of hands-on testing or an independent product review. Report a correction.
Cover: AI-generated conceptual editorial illustration; not a real application screenshot, observed output or before-and-after test.