Turn photos into illustrations with AI (diffusion model)

Runs a Stable Diffusion–based diffusion model inside your browser to redraw photos in anime, watercolour, oil and other styles. More flexible than “Illustration & painting styles”, with prompt control. Photos never leave your device.

Checking your environment…

This tool downloads an AI model (about 2100 MB) on the first run. Because of current server-side bandwidth limits, the download can take around 10 minutes; later runs use the copy stored in your browser. WebGPU is required.

Drop a file or click to choose

JPG / JPEG / PNG / WEBP / AVIF / GIF / BMP / HEIC / HEIF / SVG / TIF / TIFF

Files never leave your device

Settings

Style

Strength0.55

Number of images

Output format

Which one to use — lightweight vs. diffusion

There are two tools for turning photos into illustrations. Choose by your device and what you want. We suggest trying the lightweight one first and moving to the diffusion model if you want more.

Illustration & painting styles

Method: Lightweight GANs (AnimeGANv2 / fast-neural-style) that swap only the visual style
Requirements: Works without WebGPU, including phones and older laptops
First download: About 8MB per style, about 50MB for all 7
Time per image: Under a second with WebGPU, a few seconds on CPU
Style control: Pick from 7 fixed styles
Result: Composition, expression and facial features stay intact; rarely fails
Best for: Quick avatars, or devices with modest specs

AI illustration (diffusion)This page

Method: Diffusion model (LCM Dreamshaper v7) that repaints the photo from a prompt
Requirements: WebGPU required, 4GB+ of GPU memory (shared memory on integrated GPUs)
First download: About 2.1GB. Currently around 10 minutes due to our server's bandwidth
Time per image: 20–40 seconds on an integrated GPU, a few seconds on a discrete GPU
Style control: 5 presets plus free-form English prompts
Result: Richer, painterly results; higher strength drifts away from the original face
Best for: Gaming PCs or Apple Silicon Macs, and people who want fine control over the style

Criteria	Illustration & painting styles	AI illustration (diffusion)This page
Method	Lightweight GANs (AnimeGANv2 / fast-neural-style) that swap only the visual style	Diffusion model (LCM Dreamshaper v7) that repaints the photo from a prompt
Requirements	Works without WebGPU, including phones and older laptops	WebGPU required, 4GB+ of GPU memory (shared memory on integrated GPUs)
First download	About 8MB per style, about 50MB for all 7	About 2.1GB. Currently around 10 minutes due to our server's bandwidth
Time per image	Under a second with WebGPU, a few seconds on CPU	20–40 seconds on an integrated GPU, a few seconds on a discrete GPU
Style control	Pick from 7 fixed styles	5 presets plus free-form English prompts
Result	Composition, expression and facial features stay intact; rarely fails	Richer, painterly results; higher strength drifts away from the original face
Best for	Quick avatars, or devices with modest specs	Gaming PCs or Apple Silicon Macs, and people who want fine control over the style

No WebGPU, or want results right away? Use the lightweight version. It also keeps the composition intact.Illustration & painting styles

How to use

1
Add a photo
Drop one photo. The central square (512×512) is used, so photos with the face in the middle work best.
2
Choose a style and strength
Pick anime, watercolour, oil, comic or pixel art, or write your own English prompt. Higher strength gives a stronger style; lower keeps more of the original face.
3
Generate and pick
The requested number of images is generated side by side. Save the ones you like; running again with the same settings gives different results.

How this tool, which redraws with an image-generation AI, differs from converting into seven styles at once is explained in the article “How to Turn Photos into Illustrations and Anime Art” on our blog.

How it works and what to know first

LCM Dreamshaper v7 (MIT licence) converted to fp16 ONNX runs on onnxruntime-web with WebGPU. The photo is encoded to a latent, noise is added, and the model redraws it in 4 steps following the prompt (img2img). Strength is the amount of noise; around 0.5 changes the style while keeping the person recognisable.

The models total 2.1 GB, so the first download takes a while. Because of current server-side bandwidth it can take around 10 minutes; this will improve with an infrastructure move. Later runs use the copy stored in your browser. WebGPU and about 4 GB of GPU memory (shared memory on integrated GPUs) are required; each image takes 20–40 s on an integrated GPU or a few seconds on a discrete one.

Frequently asked questions

Are my photos sent to a server?

No. The models are downloaded into your browser and generation runs on your device’s GPU. Processing statistics used to fix bugs (device specs, error types and so on; never the photo itself, its file name or the prompt) are sent to this site's server. You can turn this off with the switch under “Help us improve the service” at the bottom of the page.

Why is the download slow?

Server-side bandwidth currently caps at about 4–5 MB/s, so 2.1 GB takes around 10 minutes. A bandwidth upgrade is planned. The download happens once; later runs use the stored copy.

Why does the face no longer look like the person?

Diffusion img2img drifts further from the photo as strength rises. Lower it to 0.4–0.5 to keep the likeness. If facial details break up, run “Face restoration” on the result.

How is this different from “Illustration & painting styles”?

That tool uses lightweight GANs that keep the composition and only change the rendering (seconds, 8 MB models). This one redraws the image with a diffusion model, so it is more flexible and prompt-driven, but the model is large and WebGPU is required.

Help us improve the service

If a conversion fails, sending us the circumstances helps us track down the cause. Knowing which devices and formats fail most often, and how long processing takes, lets us revise the supported formats and fix the bugs behind them. If you don't mind, leaving this on is a real help.

What is sent

The tool you used and the settings you chose (output format, quality, etc.)
Your device specs (GPU, memory, core count)
The input format with a rough size
How many files succeeded or failed and how long it took
The error type and details when something fails

What is never sent

The files themselves and file names
Text, subtitles or prompts you add to a video
The text to be read aloud, voice descriptions, the pronunciation dictionary and saved voices
The text in PDFs, the text you rewrote or redacted, passwords, font names and the page ranges you enter
IP addresses

The data goes only to this site's own server, never to a third-party service.

Details sent by specific tools (counts, times, kinds and yes/no only)

In addition to “What is sent”, text to speech and the PDF tools send the following details.

Text to speech: The approximate number of characters and sentences, how many parts the text was generated in, and how long the AI model took to load.
PDF tools (all): The page count range, whether editing is restricted, whether a permissions password was entered, and whether the PDF is signed or encrypted.
PDF text editing and redaction: The numbers of rewrites, deletions, white-outs and redactions, the number of rewritten characters by source (original font or fallback font, the latter as a range) and the fallback typeface, the kinds of reasons why lines cannot be edited, and the number of pages turned into images by redaction and the reasons.
PDF page operations and conversion between images and PDF: The numbers of inputs and outputs, of pages rotated, deleted, duplicated or moved, of images inserted as they are, converted, in HEIC format or with corrected orientation, and of pages rendered at a lower resolution (all as ranges), whether the original was updated in place or a new document was created, and whether the original protection was reapplied, form fields were written into the page and bookmarks were added back.
PDF compression: The numbers of inputs and outputs, of images downsized, not downsized and black-and-white, of downsized images whose transparency was also downsized, that stayed lossless or that were converted from CMYK to RGB, and how many times the file was compressed (all as ranges), the kinds of reasons images were not downsized, whether the target size was met and the kind of reason it was not met, the range of how much smaller the file got, whether the original protection was reapplied, and whether the original PDF was returned unchanged.
PDF cropping and page numbers: The numbers of inputs and outputs and of pages cropped, numbered or with their margins reduced (all as ranges), how the margins were set, how the pages to crop or number were chosen, the number format and position, the range of the starting number, that the original was updated in place, and whether the original protection was reapplied.

Sending