Render to walkthrough
video, with AI.
AI image-to-video for architecture is a class of generative system that takes a static architectural render and animates it into a short cinematic clip — a slow walkthrough, an aerial drone pan, a time-lapse of light moving across the facade. In Renovato this lives as the V.01-V.06 preset set, with each motion routing to the vendor (OpenAI Sora, Google Veo, ByteDance Seedance, Kling) that best handles it.
What image-to-video actually generates
Image-to-video starts from a single still image and generates a short clip (typically 4-10 seconds) that preserves the source frame and adds plausible motion. For architecture, the motion is almost always one of: a walkthrough (camera moves through the space), a push-in (camera approaches), a pan (camera arcs), a drone shot (overhead motion), an ambient motion (curtains move, trees sway), or a time-lapse (light shifts across the scene).
The model doesn't reconstruct a 3D scene — it's an inference, scene-aware but not geometric. A 5-second Kling clip won't be perfectly stable in a way a Lumion fly-through is; what it gives up in metric accuracy it makes up for in cinematic feel and turn-around speed.
Six motions, six routings
- V.01 Walkthrough → routed to Kling 2.6 Pro for stable long-shot camera physics and natural human-pace motion.
- V.02 Drone → routed to Veo 3 (Google Gemini) for its precision on aerial framing.
- V.03 Push-in → routed to Sora (OpenAI) for its strong narrative-frame fidelity.
- V.04 Pan → routed to Seedance (ByteDance) for photorealistic interiors.
- V.05 Ambient motion → routed to Veo 3 for subtle particle-level animation (curtains, water, trees).
- V.06 Time-lapse → routed to Sora for long-context lighting transitions.
The user picks the motion; Renovato picks the engine. The inspector lets you override the routing if you already know which vendor performs best on a particular scene.
How to ship a walkthrough
- Drop your hero source render (interior or exterior) into the atlas.
- Drag V.01 Walkthrough onto the canvas, connect the source.
- In the inspector, pick clip duration (5s default, up to 10s) and motion intensity (subtle / standard / aggressive).
- Click Generate. Cost: 5 credits. Wall-clock: 60-180 seconds — Kling is a long-poll generation. Renovato shows progress in the node; the result lands when fal finishes.
The output is a node in the atlas with a video URL, poster frame, and an explicit edge back to the source render. Drag it into the Studio timeline for cuts, transitions, and 4K MP4 export.
Stitching multiple clips into a reel
A typical client-meeting reel is 60-90 seconds and stitches 6-12 clips: walkthrough → drone → ambient → time- lapse → walkthrough → drone. Each clip is a separate preset run. The Studio editor reads all of them out of the atlas as draggable thumbnails.
Pre-wired workflow templates speed this up: the Drone Pack generates four drone passes at different angles in parallel. The Seasonal Reel generates four seasonal still variants and stitches them into a four-clip first-last-frame video loop.
Where it falls short
Image-to-video is short by nature. The longest practical clip from any current vendor is around 10 seconds; for a 30-second hero shot, plan to chain three 10s clips. The Studio editor handles the cuts; just expect some seam visibility unless the cuts hit on strong action edits.
People-in-motion remains the weakest area for all vendors. A static populated render fed into V.01 Walkthrough generally produces good camera motion but stiff or oddly-articulated figures. Reserve people-heavy clips for the ambient scenes (V.05) until the dedicated pedestrian-flow preset (V.07, on the roadmap) ships.
