remove.ac · Independent guide
How to choose a local or cloud background remover
A checklist for comparing image data paths, usable output and costs, with a source-specific Sync API example.
Published · Sources checked .
These guides are available in English. Choosing another language opens that language’s help center.
This is an independent guide from remove.ac. It is not an official remove.bg notice, and it does not present remove.ac as an official successor, migration partner or endorsed provider.
Start with two questions: where will your image be processed, and what file do you need at the end? For this checklist, “local” means that the selected processing step runs on your device; “cloud” means that a remote service performs it. A browser interface alone does not answer either question. This is a selection method, not a ranking of products.
Check the data path
If keeping image data on your device is a requirement, look for evidence covering the complete workflow: selecting a file, creating previews, processing, exporting and handling errors. Ask whether thumbnails, filenames or other image-related data are sent elsewhere, and whether a failed local run can switch to a server.
Check offline claims separately. A model or application may need an initial download. A workflow that works after that download is not necessarily available on a fresh device without a connection. Record the tested browser, device, version and steps.
For a cloud option, check retention, deletion, access to output links and the settings used by your integration. For example, Leonardo's current Sync API documentation distinguishes normal result storage from its optional ephemeral mode; that option is not available on its Async API. That is a configuration-specific statement about that API, not a general rule for cloud background removers. Leonardo Remove bg API documentation
Compare the result you can actually use
Test your own representative images. Inspect the downloaded files at their intended size, including hair, transparent objects and other edges that matter to your work. Check output dimensions, transparency, format, watermarks and any need for manual repair. A successful request is not enough if the result is unusable.
The Koutuxia WebGPU/WASM case study illustrates why output checks belong beside timing measurements. It records a single-image inference comparison on September 13, 2026 using an Apple M4 with 16 GB memory, ORT-web 1.27 and a 1024-input int8 model; the reported WebGPU output was incorrect. Its separate telemetry table is an October 14 snapshot covering September 8–October 11. Those findings concern the documented Koutuxia configurations, not later releases or remove.ac. Koutuxia WebGPU/WASM case study
| Your requirement | Evidence to request |
|---|---|
| Image data must remain on the device | A tested local mode and its complete data-flow description |
| Work must continue without a connection | The exact offline steps, prerequisites and device limitations |
| A repeatable automated workflow | A working API or CLI, versioned documentation, errors and retry behavior |
| Predictable spending | Minimum payment, included volume, expiry, extra calls and failed-call charges |
For either option, compare the total spend and effort with the number of usable results. Keep trial limits, paid production limits and one-time setup costs separate. Do not assume one processing location is always faster, cheaper or more accurate; measure the workload you need.
For an example of how to read published measurements, see Koutuxia’s browser timing data. The dataset covers Koutuxia’s browser-side processing and includes multiple historical model versions. Check the model_version and observation window in each downloadable JSON row before comparing timings. These are Koutuxia records, not remove.ac measurements or a local-versus-cloud speed ranking. Koutuxia’s browser timing data
Where remove.ac stands
remove.ac remains a local preview awaiting production acceptance. Local workflows and selected pages have been checked in specific versions; these checks do not establish every processing or data path. Public API, production authentication and payments have not been opened. Local test-account screens are separate from production authentication. CLI availability, API compatibility, pricing advantages and comprehensive no-upload behavior remain unverified.
Next step
Write down your data-handling requirement and required output, then test one representative image in a tool whose documented behavior meets both. Expand to a batch only after checking the result, failure handling and costs.
Sources
Official provider documents and original Koutuxia pages used for this draft. Provider policies and datasets may change; check them again before acting.
- Leonardo Remove bg API documentation
The ephemeral option is available only on Sync, not Async; this is not a complete deletion guarantee. Checked 2026-10-28 UTC.
- Koutuxia WebGPU/WASM case study
The September 13, 2026 single-image experiment and separate October 14 telemetry snapshot retain their historical configuration and failure. Model hashes, image IDs and full scoring/timing methods were not publicly established in this review. Checked 2026-10-28 UTC.
- Koutuxia’s browser timing data
A rolling dataset for Koutuxia browser processing. Read each JSON row’s model_version and observation window; no timings or ranks are transferred to remove.ac. Checked 2026-10-28 UTC.
Definitions, evaluation steps and project-status notes are editorial material. They do not report new product, network or API tests.
Independent editorial material from remove.ac. This page does not announce a partnership, endorsement or official-successor relationship with the providers mentioned.