Make an Old Photo Talk: 5 AI Tools Tested (2026)

The photo of my great-grandfather is 9 × 12 cm, postcard format, a silver gelatin print from the 1930s. He's wearing a stiff collar and looking at the camera with the seriousness that photographers of the era imposed on their subjects. I wanted to make him say something.

Not a deepfake. Just to hear — or imagine hearing — what that voice might have said. Hello, perhaps. Or nothing. Just to see those lips move.

That's how I spent five weeks testing lip animation tools on family photos from the 1920s to 1975.

What AI Tools Can (and Cannot) Do With Your Old Photos

A "talking photo" is a short video in which a still photo is animated by syncing the lips to an audio file. The result: the person in the photo appears to be speaking a text, a phrase, a speech.

The technology rests on two components: a video generation model that animates the face, and a lip sync system that aligns lip movement with the phonetic shape of the audio. In 2026, several tools combine both into a single pipeline.

What these tools do not do: correct flaws in a photo. If the photo is blurry, the video will be blurry. If the face is in profile, lip sync will be poor. If the photo has scratches crossing the face, they will animate with the scratches.

Photo restoration is therefore a non-negotiable prerequisite. Not optional. Prerequisite.

The 5 Tools Tested to Make an Old Photo Talk

I tested each tool on the same corpus: 12 portraits from family photos (1920–1975), black-and-white and color, variable quality. The same 15-second audio file for all.

Tool	Free trial	Price	B&W quality	Face naturalism	Recommended
D-ID Creative Reality	5 credits	From $5.99/month	Good	Excellent	✓ Yes
HeyGen	1 min trial	From $29 USD/month	Medium	Good	Partial
ElevenLabs + D-ID	ElevenLabs free	~$6/month (D-ID)	Good	Excellent	✓ Yes (combo)
Edimakor HitPaw	Watermark	From $49 USD/year	Poor	Poor on B&W	No
Filmora AI Talking	Watermark	From $49.99/year	Poor	Poor	No

D-ID Creative Reality Studio is the best-suited tool for old portraits. It was designed for business presentations with "talking avatars," but its algorithm handles silver gelatin photographs correctly. Lips don't deform abnormally. The wrinkles and strong contrasts of 1935–1955 photos don't disrupt the generation.

HeyGen delivers good results on recent photos with high resolution. On black-and-white photos before 1960, artifacts are visible: the mouth area pixelates slightly at certain syllables. Usable, but not optimal.

Edimakor and Filmora are video editing software that added a "talking photo" feature in 2024–2025. On recent high-definition photos, the results are acceptable. On old black-and-white photos, the generated lips are artificial and visibly emerge from the face. These tools are not designed for family archives.

Which Type of Photo Gets the Best Results?

Lip sync works within a very strict spectrum. What works well:

• Front-facing or slight three-quarter portrait (less than 30° to the side)
• Minimum resolution of 600 × 800 pixels after scanning or restoration
• Face occupying at least 40% of the image surface
• Even lighting without hard shadows crossing the mouth

What consistently fails:

• Pure profile (less than 50% of face visible)
• Group photo where each face is less than 2 cm on the original
• Heavily damaged photo with scratches or stains on the mouth area
• Blur on the face, even slight (generation amplifies the blur)

From my corpus of 12 portraits, 7 yielded usable results. 5 were too much in profile, too small, or too damaged. That's a decent ratio but lower than what you get with simple animation (natural movement without audio), which tolerates lower-quality photos.

The rule I followed: if the face isn't sharp and frontal enough for a talking photo, I animate without audio with Incarn — the results on old portraits are better, at $1.99 per photo.

Making a Photo Talk for Free: What Exists in 2026

The question everyone asks first. The honest answer: free with real limitations.

D-ID offers 5 trial credits at sign-up, no credit card required. Each credit corresponds to a video of up to 15 seconds. That's enough to seriously test on 5 different photos.

HeyGen gives 1 minute of free generation. More than enough for a trial, but the quota runs out quickly if you want to compare multiple photos.

ElevenLabs (for the voice) offers a free plan with 10,000 characters of text-to-speech per month, enough to generate short phrases. Combined with D-ID's 5 free credits, you can create up to 5 talking photos without spending a cent.

What the free tier doesn't do: tools like Edimakor or Filmora in trial version add a visible watermark to the entire video. Unusable for sharing with family or on social media.

No free tool delivers good results on black-and-white photos from the 1930s–1960s. They require more computing power and free models are capped. If that's your case, D-ID's paid tier ($5.99/month) remains the only serious option.

Tutorial: Make an Old Family Photo Talk in 6 Steps

Here's the method I settled on after my tests — the ElevenLabs + D-ID combination.

Step 1 — Prepare the photo. The photo must be scanned at high resolution (1200 DPI minimum for a flatbed scanner, or the equivalent with a smartphone in natural light). Restore scratches and stains with Remini or Adobe Photoshop Express before continuing. A damaged photo will produce a damaged video.

Step 2 — Create the voice on ElevenLabs. Go to elevenlabs.io. Select a voice from the library (choose by age, gender, language). Type the text you want the person to say. Export as MP3 or WAV.

Step 3 — Import to D-ID. On studio.d-id.com, create a "presenter" and import your photo as the avatar. Import the ElevenLabs audio file. Launch the generation.

Step 4 — Check the lip sync. The generated video runs for the duration of the audio. Watch the mouth area carefully. On good front-facing photos, the sync is natural on the first generation. On partial profile photos, you may need to adjust the crop angle.

Step 5 — Download as MP4. D-ID exports directly to MP4. The quality is sufficient for WhatsApp, Instagram, or embedding in a funeral slideshow or digital family album.

Step 6 — Keep the original. Never overwrite the original scanned photo. The generated video is a derivative creation, not a replacement.

Is It Legal to Make a Deceased Relative's Photo Talk?

The question arises, and it deserves a clear answer.

In most countries, the image rights of deceased persons belong to their heirs for a set period after death. For strictly private and family use — sharing the video with your own children, showing it at a family gathering, using it in a memorial tribute — the legal risk is essentially zero.

What is clearly prohibited: using a talking photo to make a deceased person say commercial or political statements, or any content they would likely have refused during their lifetime.

For photos of living people, explicit consent is required before any use. An unconsented talking photo of a living person constitutes a violation of their image rights, regardless of the tool used.

Tools like D-ID and HeyGen specify in their terms of service that the user is responsible for respecting third-party rights. No platform filters created content, but videos circulating publicly expose their author.

For a private family tribute of an ancestor who passed away several decades ago: use these tools without legal concern, limited only by good taste.

When Silent Animation Is Enough (and Why It's Different)

Making a photo talk and animating a photo are two distinct things.

Standard animation — the kind you get with tools like Incarn or Runway — creates natural face movement: eye blinking, slight head rotation, simulated breathing. No audio required. The photo comes to life without speaking words.

Animation with lip sync adds a layer of complexity: it forces movement to match precise phonemes, which imposes strict requirements on photo quality and angle.

Result: on very old or poor-quality portraits, simple animation often delivers better results than a talking photo, because the model isn't constrained to reproduce precise lip shapes.

The practical rule I adopted: if the photo is frontal, sharp, and at good resolution, try the talking photo. If the photo is in profile, small, or damaged, simple animation is more reliable. Incarn is optimized for family portraits, including black-and-white photos from the 1930s–1970s — where generalist tools fail. First animation free at sign-up, $1.99 per photo afterward.

Both approaches complement each other. The talking photo for speech and emotion. Simple animation for presence and memory.