Digital Time Machine: How to Animate Old Photos and Transport Friends into Them Using AI

The Concept of Virtual Time Travel

Modern generative models allow you to recreate the effect of traveling through time. Executing such a project does not require complex video editing skills—all it takes is a set of basic source images and properly formulated text prompts.

The video creation process is built upon three key steps:

1. Finding an archival photo of a memorable or recognizable location.

2. Restoring and colorizing the historical shot to enhance detail and atmosphere.

3. Animating the restored image using the Minimax H3 neural network while inserting modern-day people into the scene.

Location Preparation and Photo Restoration

Archival urban shots (for example, from the PastVu platform) work great for experimentation. A 1986 photograph of Kalininsky Prospect in Moscow featuring the Grauerman Maternity Hospital building—with a large screen on its facade displaying a speech by a Soviet leader—was selected as the base image.

An attempt to restore the source image using Krea 2 proved insufficient for such tasks, so the processing was carried out using the Qwen Edit 2511 model. One limitation should be noted: if there are more than four faces in the shot, the model may struggle to redraw them accurately.

Example prompt for cleaning and colorizing the photograph:

restore and colorize this photo, remove the stains, dust spots, noise, scratches and stripes from the image, fill in any gaps, ripped or torn sections, turning it into a high quality photograph, reversal film photography, remove stains and crease marks, repair gaps and fold or bends

Video Generation in Minimax H3

To seamlessly blend the historical background with real people without distorting their facial features, the initial scene description was expanded and translated into English using the GLM 5.2 model. The video generation and finalization were performed in the Wan2GP interface with three source images supplied simultaneously (the location and two portraits).

Final prompt for generating a 15-second video:

Static, locked-off camera shot. Two friends suddenly materialize in the center of the frame, looking around in astonishment and confusion as they realize they have traveled back in time to 1986.

Characters: Character 1 strictly maintains the exact facial features of @Image1. Character 2 strictly maintains the exact facial features and clothing outfit of @Image2. They stand close together, turning their heads to examine their surroundings with expressions of shock and wonder.

Background: The background strictly follows @Image3 — Kalininsky Prospect in 1986. A large outdoor screen is visible, displaying a broadcast of Mikhail Gorbachev giving a speech. Typical 1980s Soviet citizens are walking along the prospect.

Consistency constraints: It is absolutely crucial to keep all background buildings, the TV screen, and the walking pedestrians completely unchanged and identical to @Image3. The camera does not move at all. The 15-second video shows the two friends breathing and looking around while the 1986 city life continues normally around them. Cinematic lighting, photorealistic, 1986 Soviet atmosphere, temporal consistency.

Experiment Results: Pros and Cons of the Technology

As a result, a series of short clips was created where friends and relatives appear on the streets of 1980s Soviet Moscow, maintaining high portrait resemblance and emotional expressiveness.

Advantages of the Method:

• Strong emotional impact: The finished result leaves a lasting impression on participants and viewers alike.

• Natural movements: The inserted characters blend seamlessly into the environment and interact realistically with the scene.

• Minimal source assets: A single portrait photo of each person and one location image are sufficient for generation.

Limitations and Nuances:

• Loss of resemblance in motion: Sharp head turns can temporarily reduce facial recognition accuracy.

• Need for a static camera: Moving the virtual camera often distorts the original architecture of the shot.

• Animation artifacts: Background pedestrians from the archival image can sometimes move unnaturally, exhibiting a sliding or "moonwalking" effect.