How to Enhance Old Photos with AI: HDR and Upscaling Combined

You have a folder of old photos. Some are scans of prints from the early 2000s. Some are digital shots from a camera that was good for its time but maxed out at 5 megapixels. Some are downloads from an old social media account, compressed and recompressed until the JPG artifacts are part of the image. They are not ruined. They are just small and flat. And you can do something about them.

Two tools together will take you further than either one alone. This article walks through the workflow.

What "old photo" means here

Let us define the scope. This workflow is for photos that are technically intact but visually diminished. Faded contrast, flat lighting, shadows that have gone muddy, highlights that have blown to white. Low resolution from an era when 1024 pixels wide was considered large. Compression smearing from being uploaded, downloaded, and uploaded again.

It is not for photos with physical damage: tears, creases, mold spots, missing corners. The HDR enhancer and upscaler do not do inpainting. They do not reconstruct missing parts of an image. If your photo has a piece torn off, that piece will still be missing afterward. The tools improve what is there. They do not replace what is gone.

Step 1: Recover the light first

Start with HDR enhancement. The reason is simple: the upscaler works from whatever data you give it. If you upscale a photo with crushed shadows and a blown-out sky, the upscaler will faithfully reconstruct a higher-resolution version of those crushed shadows and that blown-out sky. It cannot improve dynamic range. It can only work with the tonal information present in the input.

Running HDR enhancement first gives the upscaler better material. Shadows that were solid black open up and show detail. Highlights that were flat white recover color. The midtones stretch into a wider range. The photo gains the kind of depth that the upscaler needs to do its best work.

At Free HDR Photo Enhancer, upload your photo (JPEG, PNG, or WebP, up to 5 MB), pick a Seedream model, and choose your resolution and output format. The HDR tool processes the image in about 30 to 60 seconds and gives you a watermark-free download. Seedream 5.0 Pro is the safest default for balanced results. Seedream 4.5 is worth trying if the photo is a portrait.

Keep the original. Always keep the original. The HDR result is a reconstruction, and you may want to compare it against the source later.

Step 2: Upscale the restored photo

Take the HDR output and feed it into the AI upscaler. Now the upscaler has an image with better tonal range to work with. Shadows have texture. Highlights have gradient. The model can infer finer detail because the input gives it more to infer from.

Choose your resolution. If the original was a 1024-pixel-wide photo from a 2008 point-and-shoot, 2K (2048 pixels wide) is a reasonable target. Going higher will not hurt, but it also will not add much. The upscaler can fill in plausible detail up to a point, but the gap between a 1-megapixel source and a 4K output is too wide for the model to cross convincingly. A 2x to 3x increase in linear resolution is the sweet spot.

Select the model. Seedream 5.0 Lite costs the fewest credits and outputs at up to 4K. Seedream 5.0 Pro produces finer detail at the cost of a lower maximum resolution. For most old photos, Lite is the practical choice: you want the larger output more than you want the marginal detail improvement.

Processing takes another 30 to 60 seconds. Download the result.

Step 3: Compare, keep the better version, and save

Open the original, the HDR result, and the upscaled result side by side. Zoom to full size on each one. Check three things:

First, whether the HDR pass added anything useful. Sometimes a photo already has decent dynamic range, and the HDR step is unnecessary. If the HDR version looks overcooked compared to the original, skip it and upscale directly from the original.

Second, whether the upscaled version actually looks sharper at full size. Not at thumbnail size (everything looks sharp at thumbnail size). Zoom in. Look at faces, text, edges of objects. If you see detail you could not see before, the upscaler did its job. If you see a smeary version of the same softness, the original was too small or too compressed for the model to work with.

Third, whether the combined result is better than either pass alone. Sometimes HDR plus upscale is the clear winner. Sometimes HDR alone is enough and upscaling adds little. Sometimes upscaling alone from the original is better than adding HDR first. There is no formula. Compare and decide.

Keep all versions. Disk space is cheap. You may change your mind later, or you may want a different version for a different use.

What the workflow will not fix

Motion blur stays. If the original photo is blurry from camera shake or subject movement, neither HDR enhancement nor upscaling will sharpen it. The model has no way to know which direction the blur went or what the unblurred detail looked like.

Heavy sensor noise from cheap cameras in low light confuses both tools. HDR may amplify the noise. Upscaling may treat the noise as texture and turn it into something that looks like fake film grain. Results on very noisy originals are unpredictable.

Faces at very low resolution are a gamble. An old group photo where each face is 30 pixels across might come out with recognizable facial features. Or it might come out with features that look plausible but belong to no actual person. The model reconstructs what is statistically likely. It does not know what your uncle looked like in 2003.

Extreme underexposure is not fixable. If the original is so dark that large areas are pure black, there is no data there for either tool to recover. HDR enhancement can lift shadows, but it cannot invent detail from zero.

None of this makes the workflow useless. It makes it honest. For most old digital photos, the two-pass approach of HDR followed by upscale will produce the best version of that photo you have ever seen. For the rest, at least you will know you tried.

If you want to understand more about either tool individually, we have a beginner's guide to HDR photography, a guide to AI image upscaling, and an article on why your photos look flat that covers the tonal range problem in more depth. And if you decide to run the two tools as a batch, check the pricing page to see the per-model credit costs before you start.