7 AI Prompts to Restore Old Photos (ChatGPT, Gemini, Nano Banana) — 2026

Can You Restore Old Photos With an AI Prompt?

Yes — you can restore an old photo by uploading it to ChatGPT, Google Gemini, or another AI image model and asking for a restoration in plain English. The AI will remove scratches, fix fading, sharpen detail, and even add color to black and white pictures. The catch: chat-based image models regenerate the picture rather than repairing it, so faces can subtly change. Use the prompts below to get the best possible result — and the identity-safe alternative at the end when the photo is one you truly care about.

Prompt 1: General Old Photo Restoration

The all-purpose prompt. Upload your photo, then paste:

Restore this old photograph. Remove scratches, dust, creases, and stains. Fix fading and restore natural contrast. Sharpen soft details. Do not change the people's faces, expressions, poses, clothing, or the background — keep every person exactly as they appear in the original.

The second sentence is the important one. Without an explicit instruction to preserve faces, image models will happily "improve" your grandmother into a stranger who merely resembles her.

Prompt 2: Colorize a Black and White Photo

Colorize this black and white photograph with realistic, historically accurate colors. Use natural skin tones, muted period-appropriate clothing colors, and realistic environmental colors. Keep the grain and character of the original photo. Do not alter faces, features, or composition.

AI guesses colors from context — it cannot know that a dress was actually red. If color accuracy matters, name the colors you remember: "the dress was navy blue, the car was cream white." For more on how colorization works and where it fails, see our step-by-step colorization guide.

Prompt 3: Repair Scratches, Tears, and Damage

This photo has physical damage: scratches, tears, and missing spots. Reconstruct the damaged areas so the photo looks intact, matching the surrounding textures, lighting, and grain. Repair only the damaged regions — leave every undamaged area, especially the faces, pixel-for-pixel unchanged.

Works well when damage is in backgrounds or clothing. When a tear crosses a face, the AI has to invent the missing features — results vary from impressive to uncanny, so always compare against the original.

Prompt 4: Revive Faded Colors

This color photo from the 1970s–80s has faded and shifted toward orange/yellow. Restore the original color balance: neutralize the color cast, recover natural skin tones, and bring back realistic saturation and contrast without making it look modern or over-processed. Keep all details and faces unchanged.

Old prints fade unevenly — cyan dyes die first, which is why family photos from that era look orange. Naming the decade helps the model pick the right correction.

Prompt 5: Sharpen a Blurry Old Portrait

Sharpen this soft, low-resolution portrait. Rebuild fine detail in the eyes, hair, and skin texture so it looks naturally in focus. Preserve the person's exact facial features, age, and expression — do not beautify, smooth skin, or change their likeness in any way.

The "do not beautify" clause matters: many models quietly apply modern beauty standards to faces. If the result still drifts, a dedicated deblur tool handles this better — see how to unblur a photo.

Prompt 6: Old Photo to New Photo (Full Remaster)

Remaster this old photograph so it looks like it was taken today with a modern camera: clean, sharp, natural colors, no damage or grain. Keep the people, their faces, poses, clothing, and the scene completely faithful to the original.

This is the "old photo to new photo" prompt people search for. It produces the most dramatic transformations — and the most invention, since the model rebuilds nearly every pixel. Great for sharing; think twice before treating the output as the archival copy.

Prompt 7: Identity-Lock Prompt for Precious Photos

Restore this photo conservatively. Fix only obvious defects: scratches, dust, fading, and stains. Make no creative changes. The absolute priority is that every person remains 100% identical to the original — same face geometry, same expression, same skin, same age. If unsure whether to change something, leave it as it is.

Use this when the photo is irreplaceable. It trades wow-factor for fidelity, which is usually the right trade for the only surviving photo of someone.

The Problem With Prompt-Based Restoration

Every prompt above shares one weakness, and it is worth being honest about it. ChatGPT, Gemini, and Nano Banana are generative models: they create a new image that looks like a restored version of yours. Even with strict instructions, faces can shift a few percent — a slightly different nose, a changed smile — which Reddit users describe as getting back "a stranger wearing grandma's clothes." For casual photos that is fine. For the one photo of your grandfather, it is not.

• Prompts are free and flexible — great for experiments, colorization drafts, and photos where exact likeness is not critical.
• Dedicated restoration models preserve identity — they repair the existing pixels instead of regenerating the scene.
• No prompt engineering needed — a purpose-built app applies the right restoration automatically.

If you would rather skip prompt-tuning entirely, Restore Old Photos AI – Renew restores scratches, fading, blur, and color in one tap with a model built specifically to keep faces true to the original — try it free on the App Store. For a deeper look at how app results compare, see do AI photo restoration apps really work?