40 Welcome Credits
Eligible new accounts currently receive 40 welcome credits valid for 7 days. This limited grant can cover up to three current 13-credit runs.
Image


Try another image effect without leaving the creation flow. The generator stays first, and adjacent tools are one card away.
These details sit after the working controls so the first viewport stays focused on creation. Confirm the displayed task cost again before submitting because provider prices and routing can change.
Eligible new accounts currently receive 40 welcome credits valid for 7 days. This limited grant can cover up to three current 13-credit runs.
The current low-quality GPT Image 1.5 transparent-PNG route costs 13 NanoPic credits, based on Replicate's current $0.013 price and NanoPic's $0.001-per-credit conversion.
Images made with free credits include a NanoPic watermark. Images made with purchased credits do not receive that free-credit watermark.
GPT Image 1.5 can create new assets or edit an input, but it may change identity, labels, geometry, hair, fur, or product details. Review every output before use.
A gray checkerboard can be baked into ordinary RGB pixels, and a white canvas can look transparent on a white page. Neither is transparency. This generator locks the provider request to PNG output and a transparent background. The ten launch candidates were fetched with the image/png MIME type, opened as 1024×1024 four-channel PNGs, and checked for alpha values of 0, 255, and intermediate edge pixels. That combination proves empty regions, solid subject pixels, and soft anti-aliased boundaries exist in the file itself.
The response is checked as image/png and begins with the PNG file signature. Renaming a JPEG to .png would fail this check.
Selected launch outputs contain four channels and an alpha channel. Their transparent regions use alpha 0 rather than painted white or checkerboard pixels.
Hair, fur, curved ceramic, and product edges include intermediate alpha values. Those pixels help avoid harsh stair-steps when the asset is placed on a new color.
Examples are inspected over both a warm light color and a deep blue color. The preview colors are external backgrounds; they are not embedded in the downloadable PNG.
The examples below are not white-background substitutes. Each right-hand subject is the same PNG composited over two solid preview colors after file-level alpha checks. The left side shows the source when an input image was used.
The amber bottle test checks more than the outer rectangle. The pump nozzle, ridged collar, bottle shoulders, glass highlights, base, and visible dip tube must remain readable. The transparent result can still contain generative changes, so compare the source at full size before using it for a catalog. If exact packaging, legal copy, or a barcode matters, a segmentation-only background remover is safer than a generative edit.
Try a Product Cutout
Fur creates many narrow boundaries, so a useful result needs more than a clean oval around the animal. This corgi example keeps both ears, facial expression, chest fur, paws, tail, and soft contour pixels. Inspect the silhouette on both light and dark colors: a bright halo, dark fringe, missing whiskers, or clipped ear tip will become obvious when the background changes.
Test Hair or Fur
Products often contain negative space that must become transparent too. The orange shoe test covers the gap inside the opening, lace holes, separate lace segments, sole contour, fabric edge, and complete toe and heel. A flattened background left inside any of those gaps would reveal that the cutout is incomplete. Product colors and panel shapes should also be compared with the source because GPT editing is not pixel-locked.
Check Interior Gaps
Portrait edges can fail around flyaway hair, shoulders, and light clothing. This fictional adult example is shown against two different preview colors so background spill and clipped strands are easier to spot. The result preserves an editable alpha boundary, but users should still check identity, expression, hair shape, clothing seams, hands, and skin detail before reuse. Do not treat a generative edit as an exact extraction.
Try a Portrait
Choose whether you need a new asset or a generative edit, then make the transparency requirement explicit and inspect the downloaded file rather than trusting the preview alone.
Use Text to Image for a new icon, object, mascot, logo concept, character, product mockup, or isolated design element. Use Image to Image when you already have one authorized photo and want the model to remove the scene while following preservation instructions. Upload one sharp image with the complete subject and important openings visible. Avoid screenshots, baked checkerboards, heavy motion blur, tiny subjects, and already-cropped edges.
Choose a Workflow
Name one subject, request a true transparent alpha background, and list the boundaries that matter: hair, fur, glass, lace, spokes, leaves, handles, holes, tails, thin straps, or product labels. Also say what must not appear, such as a white canvas, checkerboard, floor, shadow rectangle, border, caption, or extra object. For an uploaded source, protect identity, geometry, colors, materials, camera angle, and critical details.
Edit the Prompt
Open the finished result, download it, and confirm the response is a PNG. Place it over at least one light and one dark color. Zoom to the edge and inspect transparent gaps, partial edge pixels, fine detail, color spill, halos, stray fragments, and subject changes. If a label, face, product part, or interior opening drifted, revise one instruction and generate again. A visually attractive preview is not enough for production approval.
Generate and Verify
These pages solve adjacent but different jobs. Choose the workflow by the file you need at the end, not by the word background alone.
Use this page when the final asset must carry real alpha so it can be layered over many future designs. It supports prompt-only generation and prompt-guided editing with GPT Image 1.5.
Use the product background page when you want to keep a product visible and create a new studio, lifestyle, ad, or social scene behind it. Its output is intentionally a complete opaque composition.
Use the color-change page when the desired final image should have a specific solid or styled background rather than transparency. It is useful for profile, listing, and campaign color variants.
For exact extraction of a catalog photo without generative redraw, use a dedicated segmentation remover when available. This GPT workflow can preserve details well, but it can still reinterpret pixels.
A useful transparent PNG prompt describes the subject, full silhouette, difficult boundaries, preservation rules, and forbidden fake-background patterns in one compact instruction.
“Generate one cobalt ceramic teapot, complete handle opening and spout visible, clean isolated RGBA PNG, true transparent alpha, no floor, white canvas, checkerboard, border, text, or extra object.”
“Isolate the complete shoe. Preserve sole, panels, laces, lace holes, material, colors, camera angle, and opening. Remove every background pixel and return a true transparent PNG.”
“Preserve identity, expression, ears, whiskers, flyaway hair, individual fur edges, paws, and tail. Keep soft partial alpha without halos or color spill.”
“Keep the exact authorized mark, internal holes, thin strokes, original colors, alignment, and proportions. Return one lossless transparent PNG without redrawing text.”
The generator route is locked to background=transparent and output_format=png for Replicate's official GPT Image 1.5 model. Launch samples were fetched as image/png, opened as four-channel PNGs, and verified to contain fully transparent, fully opaque, and intermediate edge-alpha pixels. You should still inspect each personal download because model outputs can fail or contain edge artifacts.
No. A checkerboard can be painted into ordinary RGB pixels. Verify the file format and alpha channel, then place the PNG over two different solid colors. The selected examples on this page use actual transparent PNGs; their warm and blue preview colors are external inspection backgrounds.
Yes. Select Text to Image, describe one isolated asset, keep the prefilled true-alpha rules, and generate. This is useful for icons, objects, mascots, logo concepts, characters, and reusable design elements.
Yes. Select Image to Image and upload one authorized source. The model receives high input-fidelity plus preservation instructions, but it remains a generative editor and may change product labels, geometry, faces, hair, or small details. Compare the output with the source before commercial use.
The current low-quality GPT Image 1.5 route costs 13 NanoPic credits per generation, based on Replicate's current $0.013 provider price and NanoPic's $0.001-per-credit conversion. If provider pricing or routing changes, the displayed task cost should be treated as the current source of truth.
Eligible new accounts currently receive 40 welcome credits valid for 7 days. At the current 13-credit route cost, that can cover up to three generations. This is a limited welcome grant, not an unlimited plan.
Yes. Images generated with free credits include a NanoPic watermark. Images generated with purchased credits do not receive that free-credit watermark. The page does not promise watermark-free unlimited downloads.
Not always. Intermediate alpha can preserve softer boundaries, but hair, fur, glass, smoke, reflections, and translucent materials remain difficult. Inspect on light and dark colors at full size and regenerate or refine the asset when you see clipping, halos, spill, or missing detail.
No. This page creates a reusable alpha PNG with no scene behind it. The AI Product Background Generator creates an opaque studio, lifestyle, ad, or social background around a product. The final file purpose is different.
Replicate's current GPT Image 2 documentation directs transparent-PNG workflows to GPT Image 1.5. GPT Image 1.5 exposes the transparent background and PNG output controls needed by this page, so the route is explicit rather than relying on a prompt-only illusion.
Your intended use still depends on rights to the source, subject, brand, and generated content. Use only images you own or are authorized to edit, review provider and NanoPic terms, check logos and labels, and obtain any required permissions before publishing or selling an asset.
Check the complete silhouette, transparent interior gaps, fine edge alpha, hair or fur, product labels, faces, hands, laces, glass, color spill, halos, stray pixels, cropping, and any model-added detail. Then open the actual file over two contrasting solid colors.
Start with the editable prompt, choose text generation or one-image editing, and verify the alpha channel and subject details before production use.