Img2img (short for image-to-image) means giving an AI image model a picture plus a written prompt, so it makes a new picture that starts from yours instead of from nothing. A strength setting decides how far the new picture may stray from the original.
Think of tracing paper. You lay a fresh sheet over a drawing and redraw it in a new style: the shapes underneath guide your hand, but every line is new. Img2img works like that. A light touch keeps most of the original; a heavy touch keeps only its rough layout.
In a sentence: “I ran my pencil sketch through img2img with the prompt ‘watercolour harbour at dusk’, and it kept my layout but painted it properly.”
What does img2img mean?
The plain point: it is text-to-image with a starting picture added.
Hugging Face, the site where many AI models are shared, puts it simply in its guide for developers: “Image-to-image is similar to text-to-image, but in addition to a prompt, you can also pass an initial image as a starting point.”1
Most image makers are diffusion models: they normally start from random static and clean it up into a picture. Img2img skips the blank start. The same guide explains that your image is turned into the model’s internal format, some noise is added to it, and the model then cleans that noisy version up towards your prompt.1
The strength setting: how much changes
The plain point: low strength stays close to your picture; high strength lets the AI wander.
The key dial is usually called strength (this site’s prompting terms glossary calls it denoising strength). Hugging Face describes both ends: a higher value gives the model more freedom to differ from the starting image, and at the top of the scale the starting image is more or less ignored; a lower value keeps the result closer to it.1
In Hugging Face’s own software, strength runs from 0 to 1 and starts at 0.8 unless you change it.2 Other apps use their own names and scales, so check your tool’s help pages.
Img2img keeps the bones of your picture and repaints the rest. Turn strength down to keep more of the original, up to let the prompt take over.
What people use it for
- Sketch to finished image: a rough drawing sets the layout; the prompt sets the style and detail.
- Restyling: the same scene as a painting, a comic panel or a photo.
- Refining an AI image: feeding a nearly-right result back in at low strength to clean it up. The iterative workflow covers this habit.
- Chaining: Hugging Face shows image-to-image steps linked one after another, for example to apply one style and then another.1
Two cautions. Running a photo of a real person through img2img can put them in a scene they were never in; see deepfake meaning. And starting from someone else’s artwork raises ownership questions; who owns what you generate explains what is known.
Where the word comes from
The method behind it was posted in August 2021 by researchers led by Chenlin Meng, under the name SDEdit: add noise to an input image, then clean it up again.3 The short name img2img appears in the original Stable Diffusion code from the CompVis research group, which ships the feature as a script called img2img.py and credits SDEdit for the method.4
Often confused with
| Word | How it differs |
|---|---|
| Inpainting | Differs: changes only a marked area. Img2img reworks the whole picture. |
| Image-to-video | Differs: turns a still into a moving clip. Img2img makes another still. |
| ControlNet | Differs: borrows only a structure, such as a pose or outline, from a guide image. Img2img starts from the whole picture, colours and all. |
| Reference image | Differs: in many apps, a picture that steers the look or subject without being the starting point. See how to use reference images. |
Questions people ask
What is the difference between img2img and inpainting?
Img2img changes the whole picture; inpainting changes only the part you paint over. Hugging Face calls inpainting the more precise tool and image-to-image the one for sweeping changes.5
Will img2img keep my picture exactly?
No. Even at low strength the model redraws the picture, so small details can shift. That is how the method works: noise is added and then removed, as SDEdit describes.3 For a change to one spot only, use inpainting.
Does the prompt still matter in img2img?
Yes. The prompt says what to turn the picture into, and a guidance setting decides how strictly the model follows it. Hugging Face suggests pairing low strength with low guidance to stay close to the original image.1 See CFG scale meaning.
Learn more on this site
- The settings: AI prompting terms glossary, with denoising strength and other dials.
- Steering with pictures: how to use reference images.
- The engine: diffusion model meaning.
- Fixing one area: inpainting meaning.
- The words: the AI glossary.
- Related: How Does Image-to-Image AI Work? Img2img, Strength and Edits Explained.
- Related: What Is Latent Space? Latent Space Meaning in Plain English.
QUICK CHECK · 3 QUESTIONS
Pick an answer to see if you got it. Nothing is recorded.
1. What does img2img add to an ordinary text-to-image prompt?
- A starting picture for the AI to work from
- A short video clip
- A list of words to avoid
Show the answer
Answer: A starting picture for the AI to work from. Img2img gives the model your picture plus a prompt, so it starts from yours instead of from nothing.
2. What happens when you turn the strength setting up?
- The result stays closer to your original picture
- The AI has more freedom to stray from your picture
- The picture gets bigger
Show the answer
Answer: The AI has more freedom to stray from your picture. Low strength stays close to your picture; high strength lets the prompt take over.
3. How does img2img differ from inpainting?
- Img2img only changes a small marked area
- They are the same thing with different names
- Img2img reworks the whole picture, while inpainting changes only the marked part
Show the answer
Answer: Img2img reworks the whole picture, while inpainting changes only the marked part. Inpainting is the precise tool for one spot; img2img is for sweeping changes to the whole picture.
Related terms: ControlNet · Diffusion model · Image-to-video · Inpainting and outpainting · Keyframe · Lip sync
- 1Hugging Face, Image-to-image, Diffusers documentation, undated, read at source 12 Oct 2026“Image-to-image is similar to text-to-image, but in addition to a prompt, you can also pass an initial image as a starting point for the diffusion process.” ↑a“The initial image is encoded to latent space and noise is added to it. Then the latent diffusion model takes a prompt and the noisy latent image, predicts the added noise, and removes the predicted noise from the initial latent image to get the new latent image.” ↑b“a higher strength value gives the model more ‘creativity’ to generate an image that’s different from the initial image; a strength value of 1.0 means the initial image is more or less ignored” and “a lower strength value means the generated image is more similar to the initial image”. ↑c“You can also chain multiple image-to-image pipelines together to create more interesting images. This can be useful for iteratively performing style transfer on an image”. ↑d“use a combination of low strength and low guidance_scale to generate an image that resembles the initial image but is not as strictly bound to the prompt.” ↑e
- 2Hugging Face, Image-to-image (StableDiffusionImg2ImgPipeline), Diffusers API documentation, undated, read at source 12 Oct 2026“strength (float, optional, defaults to 0.8)” … “Indicates extent to transform the reference image. Must be between 0 and 1.” ↑
- 3Meng et al., SDEdit: Guided Image Synthesis and Editing with Stochastic Differential Equations, arXiv:2108.01073, submitted 2 Aug 2021, read at source 12 Oct 2026“Given an input image with user guide of any type, SDEdit first adds noise to the input, then subsequently denoises the resulting image through the SDE prior to increase its realism.” ↑a“SDEdit first adds noise to the input, then subsequently denoises the resulting image”. ↑b
- 4CompVis, stable-diffusion, GitHub README, undated, read at source 12 Oct 2026“we provide a script to perform image modification with Stable Diffusion”, run as scripts/img2img.py, “By using a diffusion-denoising mechanism as first proposed by SDEdit”. ↑
- 5Hugging Face, Inpainting, Diffusers documentation, undated, as cited on this site’s inpainting page, read at source 1 Oct 2026“You can think of inpainting as a more precise tool for making specific changes and image-to-image has a broader scope for making more sweeping changes.” ↑
Meng et al., SDEdit: Guided Image Synthesis and Editing with Stochastic Differential Equations, arXiv, 2 Aug 2021, read at source 12 Oct 2026.
The comparisons and examples on this page (the tracing paper, the pencil sketch of a harbour) are this site’s own ways of explaining, not claims from these sources.
Free printable AI cheat sheets, and AI news every week.
Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.
SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK