Enhancing Old Photos Without Losing Identity

When dealing with old photographs, there’s always that temptation to push restoration tools too far. We want higher resolution, better contrast, and cleaner lighting, but we also need to be careful about what those models actually do under the hood. Do they smooth out the skin too much? Do they start hallucinating details or changing the age of the subjects?

If you are working with upscalers or diffusion models, you need a way to tell them exactly what your boundaries are. The goal is to enhance the image without crossing into territory where identity gets lost.

The Prompt

Here’s an exact prompt in English that focuses on preserving facial integrity while still improving technical quality:

Restore and enhance this old photograph by increasing resolution, sharpness, and overall clarity. Improve contrast and lighting balance, correcting exposure and reducing noise, grain, and compression artifacts.

Strictly maintain the original facial features, identity, proportions, and expressions of all people in the image. Do not alter faces, do not beautify, do not smooth skin excessively, and do not modify age, ethnicity, or any defining characteristics. Avoid any hallucination or reconstruction that changes the original likeness.

Keep the original composition, framing, and perspective intact. Colors should be natural and faithful to the scene, avoiding oversaturation.

Why This Phrasing Matters

This prompt is crucial because it explicitly addresses common pitfalls in AI image processing. By commanding the model not to beautify or smooth skin excessively, we prevent that overly plastic, modern aesthetic that often ruins historical photos. Furthermore, by explicitly stating «Do not alter faces,» we mitigate the risk of hallucinations—those scary moments where an AI might completely change a subject’s age or expression because it thinks it knows better.

This approach is especially useful when working with advanced models like Stable Diffusion paired with ControlNet. It ensures that even as we push for higher resolution, the core identity remains untouched. If you need to adapt this prompt for specific software or hardware constraints, just let me know, and I can refine it further.