Img2img (Image-to-Image)
Img2img is an AI generation mode that uses an existing image alongside a text prompt to create a new picture while preserving the original composition.
Img2img, short for image-to-image, is an AI generation method where you supply an existing picture alongside a text prompt to guide the creation of a new image. Instead of starting from a blank canvas of random static like standard text-to-image generators, img2img uses your starting image as structural scaffolding, reinterpreting textures, lighting, subject details, or artistic style according to your written instructions.
This technique gives creators unprecedented control over layout and composition. Whether you want to turn a rough felt-tip sketch into a photorealistic architectural concept or adapt a summer product photo for an autumn seasonal campaign, img2img bridges the gap between manual drafting and automated generation.
How Image-to-Image Generation Works
Standard diffusion models create pictures by starting with a field of pure digital noise and gradually removing that noise over dozens of steps until a coherent scene appears. Img2img modifies this pipeline by introducing noise directly on top of your source picture rather than starting from absolute chaos.
The critical lever in this process is a parameter often called denoising strength or transformation strength. Measured on a scale from zero to one, it dictates how much noise the system adds to your original file:
- Low strength (0.1 to 0.3): Adds minimal noise. The output retains almost all original details, line work, and colors, making it ideal for gentle touch-ups, subtle material swaps, or slight lighting adjustments.
- Moderate strength (0.4 to 0.6): Strikes a balance between the original layout and the text prompt. The basic shapes and perspective stay intact, but surfaces, clothing, and background features transform substantially.
- High strength (0.7 to 0.9): Drowns the original image in heavy noise. The AI draws loose inspiration from your color palette and broad silhouettes, but it rebuilds facial features, props, and textures almost entirely from scratch.
Everyday Examples from High Street to Studio
Consider a small pottery studio in Leeds looking to build an online catalogue. The ceramicist photographs a simple, unglazed clay vase sitting on a workbench under flat fluorescent light. Rather than renting an expensive photography studio, they feed that plain photograph into an img2img pipeline with the prompt: "stoneware vase on a rustic oak kitchen table, soft morning sunlight, steam rising from a kettle in the background." With the denoising strength set moderately, the exact curve and proportions of the handmade pot remain untouched, while the drab background transforms into an inviting kitchen scene.
Similarly, an independent coffee shop owner in Seattle can sketch out a basic floor layout on an iPad: brown rectangles for timber booths, a green circle for a potted ficus, and blue blocks for a tile counter. Using img2img, that crude digital doodle becomes a polished 3D concept render to show contractors, saving weeks of back-and-forth blueprint drafting.
How to Use Img2img in Your Workflow
Getting dependable results from img2img requires a methodical approach. Follow these practical steps to maintain control over your output:
- Prepare a clean base image: Contrast matters. If you are using a sketch, ensure dark outlines on a clean background. If you are using a photo, crop it to the exact aspect ratio you plan to export.
- Match your prompt to the image: Describe what is already there before requesting changes. If your source image features a golden retriever sitting in grass, start your prompt with "a golden retriever sitting in grass" before appending modifications like "oil painting by an impressionist master." Neglecting to mention the existing subject often confuses the model.
- Calibrate your transformation strength: Start low, around 0.35, and test an output. Gradually dial the strength upward in small increments until the AI introduces the desired artistic flair without warping anatomy or melting straight architectural lines.
Common Mistakes to Avoid
The most frequent pitfall is setting the transformation strength too high. When newcomers crank the slider past 0.85, the AI effectively ignores the original framing, producing a disconnected image that fails to respect the original composition. If you notice fingers mutating, horizons tilting, or products morphing into unrecognizable objects, your strength setting is almost certainly too aggressive.
Another common error is submitting low-resolution source images with heavy compression artifacts. While img2img can add detail, blurry source pixels often turn into smeared, muddy textures in the final render. Feed the model crisp, well-lit inputs to get crisp, professional results.
How to configure this in Midjourney, FLUX, Ideogram, GPT Image, and ComfyUI, with side-by-side test comparisons: Denoising Strength (Image Weight) in Generator Parameters
Frequently asked questions
What is the difference between text-to-image and img2img?
Text-to-image generates a visual entirely from words, starting with pure random noise. Img2img requires both a text prompt and a reference picture, using the original image to lock down the composition, pose, and layout while altering style or details.
Does img2img keep the exact faces and shapes from my photo?
It can, provided you keep the transformation strength low. If you set the strength too high, the model will redraw facial features and geometric edges based on its training data rather than preserving your exact subject.
Can I use rough hand-drawn sketches with img2img?
Yes, turning basic line art or digital doodles into detailed concept art is one of the most popular uses of img2img. A simple color block drawing provides enough spatial guidance for the AI to understand where walls, skies, and characters belong.
