To remove a prop beside a product label, you need to keep the original lettering available while judging the cleared space. This CapCut test compares three erase selections on one candle photo and checks the label character by character before choosing the result.
The selected version removes the wooden scoop and retains the readable label. A deliberately oversized selection also removes the scoop, but leaves a blank, damaged jar. The difference is visible in the downloaded images.
The source is an AI-generated CapCut sample of a fictional candle brand, not a photograph of merchandise offered for sale. This October 21, 2026 web test covers one adjacent prop and a fully visible label. It does not reconstruct hidden packaging, test batch production, or establish marketplace approval.
The top row shows the source and selected native PNG export in order; the bottom row enlarges their labels. These are review frames from still images, not a CapCut video export or an editing-process recording.
Read the product label before selecting the prop
Read the source before generating an edit. Here the first line is MORA, the second is CEDAR, and the quantity is 180 g. The lowercase g matters as much as the larger brand letters. The wooden scoop touches the jar's right side near the rim, while the label remains unobstructed.
Keep the original as a separate image in Layers. A cleaned candidate should be compared with that source, rather than with the most recent generated version. Otherwise, a changed letter could become the reference for the next attempt.
The comparison below records characters individually. It also distinguishes a missing label from a spelling error. In the rejected output, there is no remaining label text to transcribe.
This is a visual wording check, not a claim that the generated paper texture or image pixels match the source. Fine surface texture changes are visible across the edited candidates.
Reject the result that removes part of the jar
The first rectangle included the scoop and a narrow strip of the label's right side. That output removed the scoop and retained all three lines. It remains in the comparison because touching the label did not produce a wording failure in that attempt.
A separate, deliberately broader rectangle covered the label face and extended across the scoop. In Edit tools → Select area, the selected action was Erase, followed by Apply edits. The screenshot shows what the edit was allowed to target.
The intentionally broad test selection crosses the product. The amber outline highlights the recorded region and submit control; it is an annotation on the actual interface.
The top row shows the source and rejected result in order; the bottom row enlarges the corresponding label areas. The scoop, label, and part of the jar disappear in the rejected native still export.
There is no reason to retype the brand onto this candidate: its right wall is missing, and the rim projects beyond the damaged body. Adding text would leave an altered product underneath. Return to the original instead of refining this result.
Draw the replacement selection from the source
1. Select the original image. Start in the image workspace associated with CapCut's online photo editor. This run used Create new → Image → Custom size at 1920 × 1080 from the projects page. The sample was generated with Design with AI and appeared as a separate 2048 × 1152 image. The erase tests used that full image, rather than the blank frame.
2. Open Edit tools and choose Select area. Drag a rectangle around the scoop. In this photo, its left boundary follows the space immediately outside the jar's right wall. It leaves the cream label outside the selected region and includes the tabletop around the scoop's base.
The selected attempt starts again from the same source. Its rectangle excludes the visible label and stops beside the jar; no repair is applied to the damaged candidate.
3. Choose Erase for the marked area, then Apply edits. The interface displayed 1 area ready to edit before submission. The returned image appeared as a separate item in Layers, leaving the source available for comparison.
The rectangle specifies the requested area; this test does not establish a hard pixel protection boundary. The output also has fine texture differences outside that rectangle. Judge the downloaded image rather than treating the selection outline as proof of preservation.
If you found this workflow while searching for a free AI image editor or an AI image editing tool, check your account’s current credit information before repeating the workflow. This sample used credit-based source generation. The local Erase panel did not expose a model selector or a separate per-edit quote, so this test does not establish free access or a general editing rate.
Match the letters before saving the cleaned photo
Open the candidate beside the source at a size where the small quantity line is readable. Follow the character table from the first M through the final g. Then compare the label's curved lower edge and the right wall where the scoop touched the rim.
These crops come from the final native exports. The selected label retains its wording and line order. The scoop is gone, the right wall remains visible, and the base still meets the tabletop.
In the selected image, a soft contact shadow remains beneath the jar, with no detached scoop-shaped patch at its former base. The wall-to-table boundary continues through the cleared area. At normal viewing size, the jar does not appear cut off or floating. These observations apply to this sample, not to other packaging or lighting.
All three edits start from the same source. The partial overlap attempt kept the text; the larger label-crossing attempt failed; the scoop-only attempt is the selected output. These examples do not establish a universal selection-size threshold.
For this run, Download offered PNG at 1× (2048 × 1152px). The final images were downloaded separately with Compress file off. Reopening the project retained the source and the three named candidates.
The existing inpainting guide explains local reconstruction more broadly. This test leaves general tool comparisons and background changes to those guides. Save the source with the selected PNG and its character record; for your next product photo, transcribe the visible label again before drawing an erase selection.