AI Walkthrough Video: Settings, Steps and Costs
· Updated · 6 min read
A still render gets attention; motion wins the pitch. AI video turns a single render or photo into a cinematic walkthrough — smooth camera movement, depth, and an optional classical score — without After Effects or animation experience.
Looking for the tool rather than the guide? Open the AI walkthrough video generator — upload one render, get a 5 or 10 second cinematic clip. The rest of this page is the how-to.
Why motion matters
Clients understand space far better when the camera moves through it. A 5–10 second cinematic clip makes a proposal feel finished and is perfect for Instagram and TikTok, where video massively out-performs static images.
The input: what makes a still worth animating
The clip is only ever as good as the frame you feed it, and this is where most disappointing results start. One image goes in — a finished render, a photograph of a completed building, or an AI render you just made from a sketch. What works and what doesn't is fairly predictable:
- Depth in the frame beats a flat elevation. A room shot with a foreground edge, a middle ground and a window beyond gives the model something to move through. A straight-on facade with no perspective gives it almost nothing, and the motion looks like a zoom on a postcard.
- Well-lit and high resolution. Noise and compression artefacts get amplified once the frame starts moving.
- Avoid text and fine signage. Lettering on a shopfront or a plan legend tends to warp as the camera moves. Crop it out or pick another frame.
- Use your best image, not your newest one. The one you would put first in a portfolio is the one worth spending credits on.
The actual steps, with the settings named
- Upload the still in Video Creator.
- Pick the camera move. Seven options: Auto, Zoom In, Zoom Out, Pan Left, Pan Right, Dolly In and Orbit. For architecture, Dolly In sells interior depth, Orbit sells a standalone building or a pavilion, and Pan reads best on a long facade or a linear terrace. Auto holds an almost locked-off tripod feel, which is the safest choice when the geometry is complex.
- Set the length: 5 or 10 seconds. Five is the default and the right answer for social. Ten costs 1.8x the credits and is worth it only when the move genuinely needs the extra travel.
- Set the aspect ratio: 16:9, 9:16 or 1:1. Choose it before generating, not after — cropping a finished clip throws away the framing the model composed.
- Choose a quality tier. Standard and HD are available on any paid account; the Premium and Cinematic tiers cost more credits and are worth reserving for the one clip that goes in front of a client.
- Add a score, or don't. Four built-in tracks (two classical, one ambient, one jazz) plus No Music. Silent is the right call when the clip is going into a deck you will narrate over.
- Set mood and people. Warm, cool, dramatic or calm lighting, and whether figures appear in the scene at all. Empty reads better for handover shots; people read better for retail and hospitality.
- Generate. You get an MP4 back, ready to drop into a deck, a listing portal or a Reel.
What it costs, and who can run it
A five-second clip is 100 credits; ten seconds is 135 on the default HD setting (up to 720 on Cinematic). Be aware of one hard gate before you plan a workflow around it: video sits behind the paywall. A free account cannot generate one at all — it needs an active plan or a purchased credit pack. That is deliberate; video runs on far more expensive engines than a still render. If you are evaluating Archome on free credits, evaluate the renders first and treat video as the reason to buy rather than part of the trial.
For scale: on the $29/month plan (1,000 credits) that is ten five-second clips a month, or on the $15 credit pack (500 credits), five.
Tips from actual use
- Generate the aspect ratio you will publish. A 16:9 clip cropped to 9:16 afterwards loses the composition the model built the move around.
- Slow beats dramatic. The camera moves are deliberately tuned gentle because fast motion is where inferred depth falls apart — resist the urge to ask for more.
- Generate the clip from your finished, colour-corrected render, not from a draft you plan to fix later. There is no re-render of a video at 100 credits a go.
- Silent export plus your own music track beats the built-in score whenever the clip goes into a client presentation you will talk over.
AI walkthrough vs real-time engine walkthrough
| Real-time engine (Twinmotion/Lumion class) | AI walkthrough from a still | |
|---|---|---|
| Input needed | Full 3D model with materials + camera path | One render or photo |
| Time to first clip | Hours to days | Minutes |
| Best at | Long navigable tours, exact camera control | Short cinematic mood shots for clients/social |
| Skill needed | Engine + animation experience | A sentence describing the camera move |
| Cost model | Licence + GPU + operator time | Credits per clip |
Deeper version of this comparison, with camera-move prompts that work: renders to cinematic property videos.
What about general AI video studios like Runway, Luma or Kling?
These are real, capable products, and it is worth being straightforward about what they do well. Runway's Gen-3/Gen-4 line has the most complete general editing suite of the group and leads several public benchmarks. Luma's Dream Machine (the Ray model family) is genuinely strong at image-to-video specifically, and tends to render interiors with convincing volume and believable bounced light. Kling is built for multi-shot narrative consistency — holding a character or a scene across several linked clips — which none of the architecture-specific tools here are trying to do.
The limitation, for this exact job, is that all three are general-purpose video generators trained mostly on natural photography and film footage, not on architectural drawings or CAD linework. Feed one a schematic sketch or a hard-edged plan and results are inconsistent — some produce a plausible cinematic reinterpretation, others hallucinate texture onto flat line drawings. They also have no architecture-tuned camera presets: getting a clean dolly-in on a facade or an orbit around a massing model means hand-writing a camera-move prompt and iterating, rather than picking "Dolly In" or "Orbit" from a list.
In practice the split is by input and goal. Starting from a finished photoreal render or photo and wanting a short, dependable client-facing clip in one click favours a purpose-built tool like Video Creator. Starting from a rough sketch, wanting multi-shot storytelling, or needing broader creative control beyond architecture is exactly what Runway, Luma and Kling are built for — they are worth learning if that is the actual brief.
What this does not do
Worth being precise, because the word "walkthrough" carries expectations from the real-time engines. This is a short cinematic clip generated from one still, not a navigable tour. The model infers depth from a single frame rather than reading real geometry, so the parallax is plausible rather than measured — push the camera far enough and edges will bend, which is exactly why the built-in camera moves are tuned slow and short. It will not walk a client from the entrance hall to the roof terrace, it will not hold a material spec across a long take, and it cannot be relied on for anything a client will measure. What it does, and does well, is turn your best frame into ten seconds of motion that makes a proposal feel finished. For a long navigable tour of a fully modelled building, the real-time engine in the table above is still the right tool.
Animate your render now: open Video Creator.