Perspective Warp an Image
Pick an image and set the four corner positions — straighten a skewed shot or add a tilt, applied in your browser. Nothing is uploaded.
Perspective-warp a photo, in your browser
Pick an image and set where its four corners should sit — a perspective
transform is applied with ffmpeg, entirely in your browser. Corner fields accept
pixel coordinates, plus ffmpeg's built-in W and H constants for the right and
bottom edges. Two things you'll do most:
- Straighten a skewed rectangle — a photo of a whiteboard, book page, or
building shot from an angle looks like a keystone (its edges lean in). Switch
Mode to
correct, then set the four corners to where the actual rectangle sits in the tilted photo; the tool stretches that region to fill the frame, so a leaning document comes out square. - Add a deliberate tilt — switch Mode to
distortand push the image's corners to new positions for a 3-D, laid-back, or floating-card look. The far right and bottom defaults areWandH, so identity is0,0·W,0·0,H·W,H.
Interpolation trades speed for smoothness: linear (default) is fast;
cubic resamples with smoother diagonal edges, useful for text and fine lines.
Worked example
For a 640×640 photo of a document shot slightly from the left, leave Mode on
correct and set the corners to the document's real pixel corners — say
Top-left 80, 50, Top-right 560, 50, Bottom-left 15, 610,
Bottom-right 625, 610. The tool builds the ffmpeg filter
perspective=x0=80:y0=50:x1=560:y1=50:x2=15:y2=610:x3=625:y3=610:interpolation=linear:sense=source
and the skewed page is pulled back into a clean rectangle at the same output
size. Want the opposite? Try the Inset 64px frame preset in distort mode
on a small fixture: the image is pushed inward and remapped by ffmpeg.
Correct vs distort — which mode?
- correct (
sense=source) — the four corners describe where a rectangle currently sits in your (skewed) photo, and that region is stretched to fill the whole output. Use it to remove perspective: deskew scans, flatten keystoned slides, square up a photographed sign. - distort (
sense=destination) — the four corners are where the image's own corners get pushed to. Use it to add perspective: tilt a screenshot, fake a 3-D card, make a mockup lean.
An identity setup (0,0 · W,0 · 0,H · W,H) is a no-op in either mode and
returns the image unchanged — a safe starting point.
Limits and edge cases
- Input files up to 8 MiB; any image format ffmpeg can decode works (PNG, JPEG, WebP, BMP, GIF, …). The output keeps the input's exact dimensions — the frame is never resized, only the content inside it is remapped.
- Corner coordinates accept pixel numbers from
-100000to100000, or exactlyW/Hfor the image width/height edge. Arithmetic expressions such as0.12*Ware deliberately rejected because browser ffmpeg builds do not evaluate them consistently for the perspective filter. - The four corners should form a real quadrilateral. If all numeric corners are collinear or collapse to (nearly) the same point — zero area — the run is rejected with a hint.
- Areas exposed by a
distortwarp use ffmpeg's own edge handling; there is no custom background-fill color option on this filter. - The output format matches the input. To change format, run the result through a dedicated image-convert tool afterward.
- Transparency is not preserved: processing happens in opaque planar RGB, so a transparent PNG comes back fully opaque.
FAQ
What's the difference between correct and distort mode?
In correct mode the four corners describe where a rectangle already sits in your (skewed) photo, and that region is stretched to fill the frame — this removes perspective, so a slanted document comes out square. In distort mode the four corners are where the image's own corners get pushed to, which adds perspective for a tilt or 3-D effect. Same four inputs, opposite direction.
How do I straighten a photo taken at an angle (keystone correction)?
Use correct mode and set each corner to where the true rectangle's corner
lands in your tilted photo. For a 640px-wide whiteboard shot from below, its top
edge may look narrower, so you might set Top-left to 80, 40 and
Top-right to 560, 40 while the bottom corners stay near 0,H and W,H.
The narrower top is stretched wider and the board comes out square.
Why do I enter pixels instead of percentages?
ffmpeg's perspective filter reliably accepts pixel numbers and its W/H
edge constants. Some ffmpeg builds accept arithmetic expressions in other
filters, but the browser build used here does not consistently evaluate
0.12*W-style expressions for perspective coordinates. Pixel fields are more
honest: they match what actually runs in the CLI and page.
Does warping change my image's size or resolution?
No. The output has exactly the same width and height as the input. A perspective warp only remaps where pixels land inside that fixed frame; it never crops or rescales the canvas. If you also need to crop or resize, run the result through the matching crop/resize tool.
When should I use cubic interpolation instead of linear?
linear (the default) is fast and fine for most photos. Switch to cubic when
the warp is strong or the image has crisp diagonal edges — text, line art,
UI screenshots — where cubic resampling produces smoother, less jagged edges.
It's a little slower but usually worth it for documents.
Is my photo uploaded anywhere?
No. The page downloads an ffmpeg WebAssembly build once and then processes your file locally in the browser tab — the image never leaves your device.
Developer & Automation Access
Run it from the terminal
Same engine as this page, headless — via the gizza CLI:
gizza tool image-perspective-warp 'url=https://example.com/input' 'tl_x=0' 'tl_y=0' 'tr_x=W' 'tr_y=0' 'bl_x=0' 'bl_y=H' 'br_x=W' 'br_y=H' 'mode=correct' 'interpolation=linear'New to the CLI? Get gizza →
Open it by URL
Pre-fill and auto-run this tool with query parameters — the names match the API/CLI:
https://gizza.ai/tools/image-perspective-warp/?url=https://example.com/input&tl_x=0&tl_y=0&tr_x=W&tr_y=0&bl_x=0&bl_y=H&br_x=W&br_y=H&mode=correct&interpolation=linearMachine-readable descriptor: tool.json — title + parameters JSON Schema for agents.