How to use inpainting with Stable Diffusion online
Inpaint is the technique of “painting” changes over an existing area of picture with AI, also known as Generative Fill. You can also zoom out of the image and add new information around the edges using a similar command, Outpainting.
We also provide a handy tool for instant background removal.
To use inpaint in draft / low resolution, type:
/inpaint a funny hat
This will return a quick 512×512 draft image. Note: if you started from a very high resolution image, you may experience some pixel compression or degregdation. For higher resolution inpaint, please upgrade to pro and specify the size you want, like:
/inpaint /size:768x768 a funny hat
Note: The higher resolution you go, the more detailed your prompt should be, or you’ll run into risks of “twins”, the same person or object duplicated. Please keep that in mind as the AI needs clearer direction from you when there are more pixels to fill in, and the amount of RAM the system needs. Try it out: /inpaint /size:1600×1600 /images:2 a funny hat
INPAINT BASICs video
Consider the image below. A specific region of an image is “masked”, and then the AI then fills in the blanks, seamlessly.
Inpaint the best way to fix little problems like glitched hands, unwanted text, and other minor details without changing the entire picture. Simply select the problem region, and prompt into it. A lot of NSFW Stable Diffusion guides will tell you it’s the easiest way to move pesky things, like clothing. Inpaint is also incredible for adding complexity to an image.
How to inpaint from WEB UI
The button appears under your rendered image, in the tools menu.
How to inpaint from TELEGRAM
Reply to an image with /inpaint sends a link to paint in your web browser.
You no longer have to prep your uploaded images, it just works.
Alternatively, you can reply to an image with /webui too, if you want to stay on the web instead of alternating between Telegram. It works either way.
First, render an image. Try this prompt:
/render /sampler:dpm2m /steps:75 /guidance:7.5 /portrait 8kUHD, award winning photography, light art by Daniel muller, raw, 23yo, brunette, swedish woman, happy, gorgeous face, highly detailed facial features, technicolor erotic body lighting, beautiful body proportions, sweater, high detailed skin, realistic skin texture, realistic colorful light patterns on body, real hair movement, 16k, dslr, sony a7 mark iv, (beautiful lights on naked body, silhouette), euphoric, (dynamic shadows), (erotic body lighting) [[[nsfw, Sketch, painting, cartoon, anime, animation, cgi, rendered, doll, lifeless, sad, depressing]]] <realvis20>
Five images will appear. Select the one you like by reply to it, and Reply with what you’d like to change about the image. In this case, let’s go for an outfit change without changing the face.
/inpaint a woman wearing a black silk dress, luxurious diamond necklace
Don’t worry too much about getting your prompt perfectly yet, you’ll have one more chance. You can edit this in the next step if you change your mind. Let’s try changing her outfit to something else.
A dynamic link to open the inpainting web app will immediately appear. Click on that and you’re taking to the mask painting screen.
The “Edit Inpaint” browser allows you to select the AI model on the left, expand positive and negative prompts (click refine inpaint prompt), and draw where you want your changes with a brush. Check our models page to get familiar with what these code names mean. This soon becomes second nature.
As this is a realistic image, we’ll go with URPM13-Inpainting. Only two models are shown here, but there are over a dozen now to choose from.
Last step: Refine your prompt, paint over the area with your finger or mouse cursor, and click the inpaint button. You can then close the window.
Finished inpainted images should look seamless. Inpaint is amazing!
In case you missed it, we also provide a handy tool for instant background removal.
Uploading your own photos instead of using AI images?
Yes, it’s possible, but please do not use other people’s photos without consent, distribute “deepfake” images. Many cities now have steep fines for persons selling or sharing deepfakes, so please only use this for personal curiosities.
Images around 768×768 work best, then /facelift to upscale 4x.
Wrong art style?
If the results do not look realistic or in the art style of the original image, try a different model from the selection menu.
Can I keep her face exactly the same, but…?
You mean a mustache? No? Try NSFW Dalle for such bold questions