Each Clipdrop tool has an open-source model behind a near equivalent: rembg for backgrounds, an ESRGAN-style upscaler, LaMa for erasing objects, a diffusion model for text to image. A local page that wires four or five of them together is a weekend. Most of it goes on installing models, choosing CPU or GPU, and queueing jobs so memory doesn't run out. It will do the easy images well. Hair, glass, large masks and relighting are where Clipdrop's tuned models are better, and a laptop without a GPU makes you wait.
Before you buildTry Clipdrop Free first: 20 uses a day of background removal, upscaling, cleanup and relight cost nothing. Build your own if you hit the cap or want images to stay on your machine.
Build me a local image tool page to replace Clipdrop. Requirements: - Python 3.12: a FastAPI server with one HTML page (drag and drop, plain JS, no build step). `python -m imgtools` serves http://localhost:4870. - Remove background: `rembg` with the `birefnet-general` model and a `u2net` fallback in .env. Output a transparent PNG, with a button that swaps in a colour or an uploaded image. - Cleanup: the user paints a mask on a canvas over the photo and `simple-lama-inpainting` fills it. Text remover: `easyocr` finds text boxes, the page outlines them, and the user clicks the boxes to inpaint. - Upscale: run the `upscayl-bin` binary that ships with the Upscayl app (path in .env as UPSCAYL_BIN) for 2x and 4x, and refuse inputs over 2048 px on the long side. - Text to image: optional; send the prompt to the OpenAI-compatible images endpoint in .env, and hide the tab when it is not set. - Jobs: one worker queue so two jobs never load models at once, a progress view, a before/after slider, and results saved in ./output/ as <original-name>-<tool>.png. - Models download on first use into ./models/. Pick CPU, CUDA or Apple MPS from .env. - Single user. No accounts, no telemetry, local-first: images leave the machine only for the optional text-to-image call. - Out of scope: relighting, uncrop, batch API, mobile app. - README: setup for CPU and GPU, the total size of ./models/, a timing in seconds for each tool on a 1024 px image measured on this machine, and the licence of each model to check.
Opens your agent with the prompt filled in, you press enter. Or copy it and paste it into an empty folder.
The first prompt builds the core. Run these one at a time in the same folder to grow it into your own Clipdrop.
Add a batch mode to the image tool page in this folder. Requirements: - A command `python -m imgtools batch <folder> --tool remove-bg|upscale` that processes jpg, png and webp files in a folder, writing results to <folder>/out/. - Skip files that already have an output newer than the input, and keep going after a failure. - Print one line per file with seconds taken, and a summary of done, skipped and failed. - Reuse the single worker queue so memory stays flat, and update the README.
Add an uncrop tool to the image tool page in this folder. Requirements: - The user picks a target aspect ratio or drags the canvas larger; the new area becomes the mask. - Fill it with a Stable Diffusion inpainting pipeline from `diffusers`, with the model id in .env. - Run it in the same queue as the other tools and warn about the VRAM it needs before starting. - Save the result next to the original as <name>-uncrop.png, and update the README with the model licence.
Add a product photo preset to the image tool page in this folder. Requirements: - One button runs remove background, trims transparent edges, centres the object on a square canvas with 8% padding, and adds a soft shadow. - Background colour and output size (default 1600 px) come from a small settings form. - Export PNG and JPEG, and show the before and after. - Update the README.
Speed and no setup: drop a photo into a browser tab and get a cutout, an upscale or an erased object in seconds, without installing models or owning a GPU. Pro also lifts the daily caps to 1,000 a day per tool.
Estimate · Everything runs on your own hardware. A text-to-image endpoint, if you add one, is billed per image by its provider. The build is token usage on your agent plan.
| Plan | Monthly | Annual, per month | What you get |
|---|---|---|---|
| Free | $0 | $0 | 20 uses a day each of background removal, 2x upscaler, cleanup and relight; 50 a day of text remover. Text to image, uncrop and replace background are listed without a daily number. |
| Pro | $15 | — | 1,000 uses a day per tool, high-resolution background removal, cleanup, relight, text remover and replace background, upscaler up to 16x, and no queue wait for text to image and uncrop. |
| API | custom | — | Usage-based pricing for developers; newer models are offered through Jasper's API. |
Free allows 20 uses a day of background removal, 2x upscaling, cleanup and relight, and 50 of the text remover.
Kinda. The core is a weekend build with the prompt on this page, but real gaps remain: Relight, uncrop and the tuned replace-background tool, Cleaner edges on hair, glass and fine detail than the open models give.
About $15 a month on the Pro plan, $180 a year (checked 2026-10-19).
Relight, uncrop and the tuned replace-background tool; Cleaner edges on hair, glass and fine detail than the open models give; Fast results without a local GPU or model downloads; Text to image with Clipdrop's models, unless you add an API; An account page and an API that other tools and Jasper's newer models plug into. If any of those matter to you, keep paying.
Yes: rembg (open source), Upscayl (open source), ComfyUI (open source).