Build an AI Time-Travel Video: Full “City Then & Now” Production Guide

Intro: Ever wanted a static street-corner photo to come alive and show decades of change? With AI image-to-image and image-to-video, you can. This article walks through the full City Then & Now pipeline—core scene plate, first and last frames, then one smooth time-travel clip.

I. Why Reuse a “Core Scene”?

First and last frames must match for believable motion. If layout, framing, and light differ wildly, the model struggles to bridge them.

Our approach: generate one environment-only plate (no people), then derive both keyframes from it. Same buildings, angle, and light—only “what changes” (people, era) is left for the model.

Comparison:

Method	Scene match	Transition	Best for
Two unrelated generations	Uncontrolled	Okay	Quick tests
Shared core plate	Fully controlled	Smooth	Pro video workflows

II. Text-to-Image: Core Scene Plate

Goal: a modern street corner with no people—the base for everything after.

Open Seedream 5 Lite — Text to Image on FuseAI Tools (/home/seedream/5-lite-text-to-image), paste the prompt and settings below, and generate.

{
  "prompt": "A modern city street corner on a sunny day. A red-brick wedge building on the corner with an old bookstore on the ground floor and a vintage street lamp beside it. Across the street, a glass curtain-wall office tower. Modern cars and pedestrians on the road. Centered composition, 16:9 landscape, photorealistic, ultra sharp.",
  "aspect_ratio": "16:9",
  "quality": "high"
}

Sample output:

Open core plate image

Core plate: locks layout, lens, and light for all following steps

Notes: this image defines bookstore, lamp, tower, angle, and sun direction—everything downstream keys off it.

III. Image-to-Image: First Frame (Modern + Elder)

Goal: add modern people and details for the start of the clip.

Open Seedream 5 Lite — Image to Image (/home/seedream/5-lite-image-to-image), upload the core plate from Section II as reference, then use:

{
  "prompt": "Keep building layout, camera angle, and lighting exactly the same. On the sidewalk at the corner, add a man in his 70s in a dark trench coat, back to camera, looking up at the glass tower across the street. Replace cars with modern models (Tesla, Toyota). Contemporary city mood, photoreal.",
  "image_urls": ["https://media.fuseaitools.com/image/ff7f3be0285d278bc3cc622da732c1b8_1775896723_2otxlf2i_63bb3e6ebc5248b29b1d69ab7c6cf4ee.png"],
  "aspect_ratio": "16:9",
  "quality": "high"
}

Sample output:

Open first frame

First frame: modern “now”—curiosity from the back-facing figure

Notes: the back view invites “what is he looking at?” and modern cars anchor the present.

IV. Image-to-Image: Last Frame (1950s + Young Adult)

Goal: same corner, new era—end frame for the video.

Stay on Seedream 5 Lite — Image to Image (/home/seedream/5-lite-image-to-image), upload the same core plate as in Section III, then:

{
  "prompt": "Keep layout, angle, and light direction the same. Shift the scene to the 1950s: bookstore sign in retro lettering; lamp unchanged but warmer glow; glass tower reads as brick. Add a 25-year-old man in vintage suit and hat, smiling. Replace cars with classic 1950s (e.g. Chevrolet Bel Air). Warm yellow grade, film texture.",
  "image_urls": ["https://media.fuseaitools.com/image/ff7f3be0285d278bc3cc622da732c1b8_1775896723_2otxlf2i_63bb3e6ebc5248b29b1d69ab7c6cf4ee.png"],
  "aspect_ratio": "16:9",
  "quality": "high"
}

Sample output:

Open last frame

Last frame: retro mood; geometry matches the first frame

Notes: silhouette, lamp position, and POV align with Section III—only era, wardrobe, and cars change. That gives I2V a clean anchor.

V. Image-to-Video: The Transition

Goal: feed first and last frames so the model interpolates from modern to retro.

Open Seedance v1 Lite — Image to Video (/home/seedance/v1-lite-image-to-video), set start image and end image, then align prompt and settings with your API/UI (resolution, duration, fixed camera, etc.).

{
  "prompt": "Fixed camera on the same corner. Opens in the modern city with an elderly man at the curb. Light warms; modern cars morph into classics; glass tower becomes brick; bookstore sign turns retro. The man grows younger—hair darkens, trench becomes a vintage suit. Ends in the 1950s with the young man smiling into the distance. Smooth, cinematic, film-like color.",
  "image_url": "https://media.fuseaitools.com/image/78121a330dc5a4bf20ee54ab02e89af7_1775901058_a1uqp9d9_300bdd84acc141c5ae50e9a9b0e1977f.png",
  "end_image_url": "https://media.fuseaitools.com/image/c5ac3f2a3efa7df1d8fed07f9a1ef34d_1775901232_x2wm1dpv_81e3ca1d9859444f8ec43b9f0cbcb426.png",
  "resolution": "1080p",
  "duration": "10",
  "camera_fixed": true,
  "seed": -1
}

Sample output:

Open sample MP4

Sample: modern-to-retro time travel in one take

Quality snapshot:

Dimension	Rating	Comment
Transition smoothness	⭐⭐⭐⭐⭐	Aligned geometry and light
Era shift	⭐⭐⭐⭐⭐	Cars, façades, wardrobe evolve naturally
Character morph	⭐⭐⭐⭐	Elder→youth reads well; slight dreamlike feel
Emotion	⭐⭐⭐⭐⭐	Nostalgic “rewind time” beat

Good fits: city promos, heritage brand stories, before/after place marketing, nostalgic short-form.

VI. Why This Pipeline Works

Strength	Why it matters
Consistency	Both keyframes inherit one plate—framing and light lock
Control	Tune each step without breaking the others
Efficiency	Generate the plate once; reuse for many variants
Natural motion	The model only models change, not guessing a new set
Extensible	Same plate → seasons, day/night, style passes, etc.

VII. More Ideas From One Plate

Reuse the same core image for other stories:

Variation	Prompt cues	Use case
Seasons	“Winter snow, coats, gray sky”	Four-season campaign
Day / night	“Neon at night, moonlight on wet asphalt”	Day vs night city edit
Style	“Studio Ghibli-inspired, warm and painterly”	Stylized short
Character	“Couple hugging at the corner”	Emotional spot

VIII. Wrap-Up

Core plate → first frame → last frame → video is a repeatable way to get pro-looking time-travel shorts: lock what stays fixed, then let the model handle change.

City Then & Now is one template—the same corner can spin into seasons, day/night, or style studies. Experiment on Seedream 5 Lite (Text to Image), Image to Image, and Seedance v1 Lite (Image to Video).