What Photos Do You Need for an AI Walkthrough Video? 7 Rules for Smooth Results

You need exactly two photos to make an AI walkthrough video: a start frame and an end frame. Which two photos you choose — and how they relate to each other — matters more than any setting you pick afterward. Good photos for an AI walkthrough video share a space, overlap in what they show, and match in light. A bad pair produces warped walls and furniture that morphs mid-shot.

This guide explains how the start-frame/end-frame workflow actually works, then walks through seven concrete rules for choosing or shooting photo pairs that generate smooth, believable camera motion. None of them require special equipment — a recent phone is enough.

How an AI Walkthrough Video Works: Start Frame and End Frame

An AI walkthrough video is generated from two still images. The start frame becomes the first frame of the finished clip, the end frame becomes the last, and the model synthesizes every frame in between — inferring a camera path through the space and rendering the move as continuous, cinematic motion.

The model doesn’t have a floor plan. Everything it knows about your space comes from visual cues in the two photos: the angle of the walls, the size of the furniture, the way light falls across the floor. From those cues it estimates depth and geometry, then generates the parallax — nearby objects sliding past faster than distant ones — that makes the motion read as a real camera moving through a real room.

That’s why photo choice is the biggest quality lever you control. The AI can only interpolate what the two frames imply. If the frames agree about the space, the motion is smooth. If they contradict each other — different light, different height, no shared geometry — the model has to invent a transition, and invented transitions are where warping happens.

In practice, the workflow is short. In Spacely, you upload a start frame and an end frame, choose one of 10+ camera presets — dolly, orbit, pan, and others — and get a finished clip back in two to five minutes, with no 3D modeling, filming, or editing. The preset shapes the path; the photos define the space. We cover picking the right move for each room type in our guide to camera movements for real estate video.

The 7 Rules for Photos for an AI Walkthrough Video

Seven factors decide whether a photo pair produces smooth motion: same or connected space, overlapping fields of view, consistent lighting and white balance, wide framing, matched camera height, minimal clutter, and adequate resolution. Get these right and most pairs generate a clean walkthrough on the first attempt.

Rule 1: Use the Same Room or Connected Spaces

The two frames should show the same room from different positions, or two spaces that physically connect — living room to kitchen through an open doorway, hallway into primary bedroom. The model needs a plausible route a camera could actually travel, so give it one.

A pair like “front of the living room” and “back of the living room” is the easiest win. A pair like “kitchen” and “upstairs bathroom” forces the model to hallucinate everything between two unrelated spaces, and the middle of the clip will show it. To feature disconnected rooms, generate one clip per room and cut them together afterward.

Rule 2: Keep Overlapping Fields of View

Some portion of the scene should be visible in both photos — a window, a sofa, a doorway, a kitchen island. Shared landmarks anchor the interpolation: the model tracks how they shift between the frames and builds the camera path around that movement.

Aim for noticeable overlap; around a third of the frame is a comfortable target. If nothing in the start frame reappears in the end frame, the model is bridging two unknowns and will fill the gap with guesses. This is also what makes doorway transitions work: keep the doorway itself visible in both shots so the path through it is implied.

Rule 3: Match Lighting and White Balance

Shoot both photos in the same session, with the same lights on and the same window conditions. The model treats lighting as part of the scene — if the start frame is warm afternoon sun and the end frame is cool overhead LED, it has to repaint the entire room mid-move, which reads as flicker and drifting color.

Auto white balance is the sneaky culprit here: two shots taken minutes apart can render the same wall in different tones. Lock white balance if your camera allows it, or apply the same edit to both files. If your photos were professionally retouched, pick both frames from the same delivered batch.

Rule 4: Frame Wide Enough to Show the Room

Wide shots that include floor, walls, and ceiling context give the model the geometry it needs to estimate depth. Tight detail crops — a countertop, a mantel, a styled shelf — leave nothing to move through, and motion generated from them tends to feel flat or distorted.

Standard wide-angle interior photography is ideal input, which is one reason listing photos convert so well. One caution: avoid extreme fisheye distortion. Lines that bow in the photo will bow in every frame of the video, and the effect compounds once the camera starts moving.

Rule 5: Keep Camera Height Consistent

Shoot both frames from the same height — roughly chest height is the interior-photography standard and reads as natural eye-level movement in the finished clip. If the start frame is chest height and the end frame is a high angle from a staircase, the interpolated camera appears to swoop or sink for no reason.

This is the easiest rule to break when mixing photos from different shoots or different photographers. When in doubt, compare where the ceiling line sits in each frame; if it cuts across the two photos at very different heights, reshoot one of them.

Rule 6: Declutter Before You Shoot

Small movable objects — cords, remotes, papers, pet toys — are hard for a model to track between frames, especially when they appear in one photo and not the other. A tidy room gives the AI fewer small shapes to reconcile, which directly means fewer artifacts.

You don’t need staging-level perfection. Clear the surfaces, straighten the cushions, and make sure anything visible in both frames sits in the same place in both. If someone moved a chair between shots, the video will show a chair that slides or duplicates itself.

Rule 7: Meet the Resolution Minimums

Use original-resolution files — roughly 1080p-equivalent (about 1,920 pixels on the long edge) is a comfortable floor, and higher is better. The model can only render detail that exists in the input, so soft, compressed, or upscaled photos produce soft, smeary video.

The most common failure here isn’t the camera — it’s the delivery path. Screenshots, photos sent through messaging apps, and images re-downloaded from a listing portal have usually been recompressed at least once. Go back to the originals from your photographer or your camera roll whenever you can.

Picking Photos for an AI Walkthrough Video From Shots You Already Have

You usually don’t need a new shoot. Professional listing and portfolio photos already satisfy most of the seven rules — they’re wide, evenly lit, decluttered, and shot at a consistent height in a single session. The real job is pairing: find two frames of the same space with overlapping views.

Group your photos by room, then look for pairs taken from opposite or adjacent corners with a shared landmark between them. A typical listing set yields several usable pairs without any reshooting. Agents can walk the full process end to end in our guide to turning listing photos into a walkthrough video; designers and stagers apply the same pairing logic to project photography, as covered in our piece on interior design portfolio videos.

Order matters too. The start frame sets where the move begins, so a pair that opens wide at the entry and ends deeper in the room mimics how a visitor actually experiences the space. Swapping the same two photos reverses the move — a pull-back reveal instead of a push-in — so one good pair effectively gives you two shots.

AI Video Photo Requirements: A Pre-Upload Checklist

Before you upload, check the pair against seven quick questions: same or connected space, shared landmarks, one lighting condition, wide framing, matched height, tidy surfaces, and original-resolution files. Thirty seconds of checking prevents most regenerations — which matters on any credit-based tool.

• Space: same room, or two spaces connected by a visible path
• Overlap: at least one clear landmark appears in both frames
• Light: same session, same lights on, same white balance
• Framing: wide shot with floor and ceiling context
• Height: both frames shot at roughly chest height
• Clutter: surfaces cleared; nothing moved between the two shots
• Files: original exports, not screenshots or portal re-downloads

The pattern behind all seven rules is the same: the AI is reverse-engineering a space from two glimpses of it. Every rule reduces the contradictions between those glimpses. Feed it two frames that agree about the room — its shape, its light, its contents — and the model rewards you with motion that looks like a camera operator walked it for real.