Venturetwins makes a Kling property tour from 8 listing photos

• Kling 4 generated a roughly 30-second property tour in one pass, according to venturetwins.
• A seven-shot script coordinated room references, camera movement and dialogue in the published prompt.
• The test used Kling's built-in speech, which venturetwins specifically wanted to try.

The Zillow listing is a 1926 house in San Francisco's West Portal, with 1,421 square feet. Kling even preserved its unusual dining-room light fixture, according to venturetwins' follow-up.

Eight uploaded listing photos supplied the visual references, with no realtor image provided, venturetwins explained. Kling had to generate the presenter across the different locations.

Venturetwins called it the first model they had tried that maintained full consistency with the reference images.

Venturetwins asked GPT-6 Astra to select listing images and align dialogue with a timestamped shooting script, the demo's most reusable artifact. The requested format was vertical 9:16.

1. 0:00 to 0:07, exterior: The realtor welcomes viewers beside the front steps and introduces the address.
2. 0:07 to 0:11, living room: She gestures toward the custom built-ins.
3. 0:11 to 0:14, dining room: Her dialogue continues the previous sentence as the camera moves sideways beside the table.
4. 0:14 to 0:17, kitchen: She describes the nearby breakfast nook.
5. 0:17 to 0:21, bedroom and bathroom cutaways: Three interior views play without a presenter while her voice describes the bedrooms and bathrooms.
6. 0:21 to 0:25, downstairs flex room: She returns on camera and mentions the bonus space and two-car garage. The prompt explicitly forbids inventing a garage view.
7. 0:25 to 0:30, backyard: She closes with the fenced garden and “Come tour today!”

One wrinkle in the published script: Image 7 is assigned to both the bedroom/bathroom cutaways and the downstairs flex room.

The presenter was specified entirely in text. Venturetwins' prompt makes her placement unusually concrete:

• Identity: The same young adult Asian woman throughout, with dark hair, a beige suit, a white top and a consistent conversational voice.
• Scale and depth: She stands on the actual floor beside recognizable furniture. Nearby furniture can overlap her lower body to establish depth.
• Framing: Eye-level camera, approximately mid-thigh upward, with her eyes near the upper third and comfortable headroom.
• Movement: Short physical sideways tracks or gentle approaches, with changing perspective between foreground and background.
• Architecture: Preserve room proportions, finishes and furniture. Keep separate reference photographs from becoming a fabricated combined room.

Kling 4.0 is available in the Artlist AI Toolkit, which lists output up to 1080p. Artlist's model guide distinguishes two controls:

• Omni multi-reference: Up to 15 references in one generation.
• Middle Frames: A start image, up to eight middle frames and an end frame to steer movement, shot changes and reveals.

The audio instructions in the prompt require exact dialogue, no music or extra lines, and each sentence finishing before the next begins. They also spell out the address pronunciation: “one forty-six.”

Venturetwins suggested a separate voice workflow: generate the complete audio track with Eleven v4, then upload it to Kling as an audio reference. That was a proposed alternative; the posted test used Kling's own speech.