
Cloudinary's documented Generative Remove effect accepts non-transparent images. A transparent product cutout therefore falls outside that input requirement, even if an ordinary opaque photograph works. Inspect the actual source and supported environment before refining the prompt or assuming that another retry will change eligibility.
Key Takeaways
- Transparent input is a documented capability limit.
- Animated and fetched images are also excluded by the cited guide.
- A processing response and an unsupported source need different investigation.
Check the real input rather than its filename
Keep the approved cutout and its original photograph separately. Inspect whether the uploaded image actually contains transparency and whether it is a still image. The extension alone cannot answer those questions.
The Cloudinary Generative Remove guide explicitly limits the effect to non-transparent images and says animated and fetched images are unsupported. Record which actual source route the failing request uses before attributing the failure to the object description.
Do not keep uploading copies of the same unsupported representation merely because the preview tile looks like a normal product picture.
Choose a supported source without sacrificing the cutout
If an authorized opaque source photograph exists, evaluate whether it is the appropriate starting point for the removal task. Keep the transparent deliverable untouched. An opaque edit and a reusable alpha cutout serve different purposes.
This is a source-selection recommendation, not a claim that flattening onto any arbitrary background guarantees a successful result. The documentation establishes the non-transparent requirement; it does not guarantee detection of your target or preservation of every product edge.
The separate cutout-and-shadow guide helps record those delivery roles. A convenient opaque preview should not silently replace the editing master.
Inspect other documented boundaries
The captured official page lists regional availability restrictions, including an Asia Pacific data-center limitation. Check the actual product environment and current provider guidance; the viewer's location is not evidence of which environment processes the asset.
It also distinguishes preparation states: a derived version can return 423 while being generated, while incoming processing can report 420 with pending status. Those documented states do not establish that every failure is temporary or that an unsupported input will eventually work.
Cloudinary further warns that very small or very large objects may not be detected, advises against faces or hands, and describes internal downscaling of large images. An accepted input is therefore not a promise of an accurate reconstruction.
Review a supported candidate as a new image
If a supported source and workflow are chosen, inspect the actual result against the retained original. Check product boundaries, background structure, retained lettering and any shadow. The controlled-removal guide covers collateral changes, while the shadow-repair guide addresses lighting consistency.
Accept only a suitable source, supported route and reviewed output. If the requirement still depends on an unsupported transparent input, keep that limitation explicit and choose another appropriate workflow. No request, environment inspection or generated result is claimed here.
Related tasks
- Remove objects from a photo with a controlled AI selection
- Should a Product Cutout and Its Ground Shadow Be Delivered Separately?
- Why Does My Product Photo Have Two Conflicting Shadows?
Sources and scope
- Cloudinary: Generative remove: Prompt scope, multiple instances, regions, remove-shadow and text-based object targeting control requested removal. The effect accepts non-transparent images, excludes animated/fetched inputs and has documented regional and processing limits; outputs are generated rather than recovered factual pixels.
Documentation-based guidance from the cited official pages captured on October 25, 2026. Scenarios and acceptance procedures are proposed, not executed tests. No account operation, measured demand, AI-citation frequency, processing benchmark or search-indexing outcome 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.