Drop images here or click to upload
PNG, JPG, WebP supported — multiple files OK
Loading AI model... 0%
First use downloads AI model (~40MB). Subsequent uses are instant (cached).
Start with a cutout task, not just an upload.
The entry should already tell you what image fits the model, where proof will come from, and how the result will decide the next export or continuation path.
Start with the real subject image you need to use next, not a collage or distant scene. Tighter framing gives the removal workflow a stronger opening position.
Best fit: one subject, clear edge contrast, and a simpler background. These files are most likely to survive into resize or compress without another removal pass.
The model can produce a clean cutout, not a guarantee that every hair, shadow, or low-contrast edge survived. Proof comes from the compare slider and the next-step join.
Keep the image flow continuous from upload intent to export decision.
Phase 28 joins image entry, proof expectation, benchmark choice, result explanation, and continuation into one asset workflow.
The entry sets the job: upload the real subject image, expect edge proof, and understand that cutout quality depends on framing and contrast.
The result should choose clean subject, hard-edge recovery, or export readiness before it recommends retry, export, resize, or compress.
The next action should feel like the same image job continuing from edge proof, not a generic image-tools prompt.
Measure whether the image flow actually stays connected.
Phase 28 P1 keeps the cross-suite language consistent while preserving the image-specific proof boundary: entry signal, proof signal, benchmark signal, continuation signal, and saved signal.
The image route should keep using the same shared sequence while still saying that cutout quality depends on framing, contrast, and edge inspection.
Track whether next_step_click moves toward retry, export, resize, or compress after the benchmark explains the cutout result.
Save and share should carry the edge benchmark and export reason, not just a generic background-removed success state.
Upload an image to remove its background
Loading AI model...
First time may take 30-60 seconds
Drag the divider to inspect hair, edges, shadows, and small gaps before downloading. On phones, swipe the handle left or right.
See whether this cutout can survive the real handoff through reusable proof assets.
These proof slots turn background removal into reusable assets: a before/after anchor, a hard-case anchor, and a route-join asset that can keep accumulating without changing the image product language.
Use the compare slider on hair, shadows, transparent gaps, and thin outlines before you treat the result as final.
Trust this anchor when the cutout still looks credible against both transparent and solid backgrounds.
Only continue into resize or compress when the before / after anchor still holds under inspection.
Low contrast, complex edges, and multi-subject frames are the most common reasons a cutout still needs retry or fallback export.
Use this anchor when the current file is closer to a difficult image than a clean one-shot cutout.
This anchor is what makes crop, retry, or fallback export a credible move instead of a generic alternate CTA.
The real question is not just whether the background disappeared, but whether the cleaned asset is credible enough to move into export, resize, or compression.
Use this asset to decide whether the image is ready for a production handoff or still belongs in removal cleanup.
The next step should feel earned by edge proof and hard-case reading, not by a generic image-tool prompt.
Match this cutout to an image-specific benchmark before export.
Phase 27 turns background removal proof into route-specific behavior: clean subject, hard-edge recovery, or export readiness.
Use this benchmark when one subject, clear contrast, and a simple background make the cutout likely to survive resize or compression.
Use this benchmark when hair, shadows, transparent gaps, or low contrast mean crop, retry, or fallback export should come before continuation.
Use this benchmark when the same edge proof holds on transparent and solid backgrounds, so the result can move into download, resize, or compress.
Turn each image benchmark into trust, action, and saved-job evidence.
Shared benchmark language stays consistent across the flagship suite: benchmark pattern -> trust signal -> action path -> saved posture.
Log whether the result matched clean subject, hard-edge recovery, or export readiness before recommending an image continuation.
Map the benchmark to retry, export, resize, or compress so the image suite moves only when the cutout quality earned it.
Keep the local saved job tied to the image benchmark and share summary, not just the download-ready state.
Run one image to see how usable the cutout is.
This panel explains what the edge quality means, when to retry, and which next route fits the cleaned image best.
A tighter crop or cleaner source usually matters more than rerunning the same weak file unchanged.
Resize or compress the cleaned asset only after you trust the edges against both transparent and solid backgrounds.
Trust the cutout when the compare slider still holds up on hair, shadows, and edges against both transparent and solid backgrounds.
Switch to crop, resize fallback, or a solid-background export when the edge problems are still visible after inspection.
Keep the free path complete, and leave the upgrade boundary honest.
The free product path should already solve the local job: remove the background, inspect proof, save/share the summary, then continue into the next image step.
Local removal, before/after proof, saved jobs, and route-to-route continuation stay available without blocking on an upgrade.
If premium depth grows later, it should extend real tasks like reusable export packs, denser proof summaries, or larger recovery flows instead of gating the current result.
Batch removal is ready for the next image step.
Download the cleaned PNG files first, then move into resize or compression only after the outputs look stable enough to keep.
Return to the image workflow with the next step already chosen.
These cards keep only local result metadata. Original images and cleaned exports are not stored here, so reruns still require re-upload.
Saved in `localStorage` on this browser only, never uploaded, never synced, and auto-cleared after 30 days.
No source images, no PNG/JPEG/WebP blobs, and no long-term batch payloads. Only summaries, saved posture, and continuation links stay behind.
Resume detector, PDF, and image jobs from one local-first task home.
This cross-suite job home keeps recent and saved work together so you know what to continue now, what to keep pinned, and what summary already travels outside the page.
Everything here stays in `localStorage` on this browser only. Nothing syncs to an account, server, team, workspace, or API.
Recent jobs surface what you touched last. Saved jobs pin the items you expect to revisit later. Each card keeps the primary path first and the alternate path second.
Shared summaries travel outside the app. Saved jobs stay here as your local return point with the proof memory that explained why the next step was trusted or switched. Clear or delete any item when you no longer want it kept for up to 30 days.
This stays free and local-first today. If real before/after evidence, bookmark, email, sync, account-level memory, export packs, or shared handoff layers come later, they should extend saved proof context without blocking the free local job now.
Something went wrong. Please try again.
Run this tool in three short steps.
上傳您的圖片
拖放或選擇一張或多張 PNG、JPEG 或 WebP 圖片。多個檔案會依序處理。
AI 移除背景
RMBG-1.4 AI 模型透過 WebAssembly 在瀏覽器中運行。不會將資料傳送到任何伺服器。
下載結果
選擇透明、白色、黑色或任意自訂顏色背景，然後下載為 PNG、JPEG 或 WebP。
What people ask before they use this tool.
AI 去背是如何運作的？
可以同時為多張圖片去背嗎？
有檔案大小限制嗎？
輸出格式是什麼？
為什麼第一張圖片處理較慢？
我的圖片資料是否安全？
去背後可以選擇背景顏色嗎？
有哪些下載格式？
去背的準確度如何？
Continue the workflow
Need to work with text too?
Extract, rewrite, or humanize text from your images.
Coda One 的背景移除工具使用完全在瀏覽器中透過 WebAssembly 運行的 RMBG-1.4 AI 模型。上傳任何圖片即可立即獲得透明 PNG。選擇自訂背景顏色，下載為 PNG、JPEG 或 WebP。無需上傳、無伺服器處理。免費且無限制。