AI Image Transformer

AI Image to Image takes an existing photo or illustration as input, combines it with a text prompt, and outputs a new image that retains the original structure while applying the style or changes you described. Unlike generating from scratch, the result always maintains some relationship to the source — making this tool useful when you want to restyle existing material rather than start from nothing.

How transformation strength controls how much of the original survives

Strength (0–1) is the most critical parameter in image-to-image work, directly controlling how similar the output is to the source:

• 0.2–0.4: Most structure and color from the original are preserved; the prompt adds only surface-level style — useful for applying a watercolor texture without changing the composition
• 0.5–0.6: A balanced point where composition largely holds but tone and details shift toward the prompt
• 0.7–0.8: Significant transformation; the source acts mainly as a structural reference while style changes substantially
• 0.9–1.0: Near-complete regeneration from the prompt; the source provides only a rough structural outline

When trying a style description for the first time, start at 0.6 and adjust toward either end once you confirm the prompt is working as intended.

Some models ignore the strength slider

Certain models — particularly those that accept reference images rather than applying strength-based blending — do not use the strength parameter at all. For these, the slider either doesn't appear or has no effect. If the model description doesn't mention "strength" support, strength adjustments won't change the output.

How output size is determined

Two sizing modes exist depending on the model: inherit mode automatically uses the source image's aspect ratio, capping the longest edge at 6,000 pixels; explicit mode requires you to choose an output resolution (128–8192 pixels) and the model ignores the source image dimensions. The mode is determined by which model you select, not a setting you can switch manually. If you upload a 3000×2000 image but want a 1024×1024 output, crop to a square first, then choose a model using explicit mode.

Using additional reference images

Some image-to-image models support uploading extra reference images (up to 15), which provide additional style or content guidance while the source image continues to supply the main structure. For example: the source is a product photo, the reference images are samples of a specific design style, and the output renders the product in that style. Uploading many contradictory reference images typically reduces result consistency.