Scale model (Seedream 5)
Image 1Scale model (Seedream 5)
Original design
Image 2Original design
To create a video like this, we have two options: a free method with results that aren't guaranteed, and a paid, professional method with more accurate results. In both cases we use the same two reference images and the same prompt, only changing the site and model used.
01

The Prompt Used

First we need two images: the scale model image and the original design image. We generate the scale model from the design image using the following prompt (the image above was generated with Seedream 5):

Scale model image prompt
Transform the building in the reference image into a physical architectural scale model made of light basswood / balsa wood, matte natural wood finish, precise laser-cut edges, thin wooden mullions for the facade glazing. Keep the exact massing, proportions and facade rhythm of the design. Place it on a clean light-grey site base with simple white/wood block context buildings on both sides, a wide empty plaza in front, small white abstract human figures and white model trees. Front-facing eye-level camera, low angle, centered composition, building in the middle third of the frame. Soft studio lighting, seamless light-grey gradient backdrop, shallow depth of field, photorealistic product photography.

Now we have the scale model (Image 1) and the original design (Image 2). Then we use the video prompt below: the site starts empty, and a hand enters from the top of the frame to place the wooden volumes floor by floor, while the camera pulls back almost imperceptibly. Once the model is complete, the pull-back accelerates, and at 8.5s the wood turns into real materials and the grey backdrop into sky, as the camera settles on the real building matching the design image.

Full Video Prompt
Vertical 9:16, 10 seconds. Use Image 1 as the exact reference for the architectural scale model (massing, number of volumes, proportions, facade, basswood material, site context). Use Image 2 as the exact reference for the real built building and its surroundings.

Main action: a stop-motion style assembly of the basswood scale model from Image 1, on a light-grey studio base, seamless grey backdrop, soft studio lighting. The site starts empty, with only the context buildings and the plaza. A realistic human hand enters from the top of the frame holding the first wooden volume (the ground floor), lowers it slowly and places it precisely on the site, presses it gently so it settles, then leaves the frame. The hand comes back with the next volume and stacks it on top, then again with the next one, placing the volumes one by one, floor by floor, until the model is fully assembled and matches Image 1. Each placement is careful, deliberate and physically realistic, with natural finger grip, real weight and a slight settling motion. Every volume is always carried and placed by the hand; nothing appears or floats by itself. When the model is complete, small white figures and white model trees pop up around the plaza.

Camera: one continuous smooth dolly-out (pull-back zoom out) for the whole video, front-facing, low eye-level, centered on the building, no cuts, no shake. Speed ramp (time remapping): extremely slow during the assembly, then progressively accelerating with an exponential ease-in curve.

0–7s: the hand assembles the model piece by piece while the camera pulls back almost imperceptibly.
7–8.5s: the hand is gone and the pull-back accelerates faster and faster, with subtle motion blur at the frame edges.
8.5s (climax): at peak speed, the scale model seamlessly transforms into the real building from Image 2: basswood becomes real materials, the grey backdrop becomes a real sky, white figures become real people, model trees become real trees, studio light becomes natural daylight.
8.5–10s: the camera decelerates smoothly and settles on the real building, matching Image 2.

Keep the building's massing and facade identical between model and reality. Photorealistic, cinematic, smooth motion.

Avoid: camera shake, cuts, jump cuts, deformed hands, extra fingers, floating volumes, volumes appearing by themselves, building distortion, warping or morphing geometry, flicker, text, watermark.
02

Choosing a Method

Whichever site you use, the steps are usually similar: upload the images as Reference Images, add the prompt, choose the model, set the settings that suit you, then click Generate.

Free, Not Guaranteed

Google Flow

Google Flow is completely free, using the Omni 1.1 Flash model.

Paid, Professional

Seedance 2.5

A professional model available on many sites. Here are two sites where you can get a discount through my links. Whichever you choose, the engine is the same, Seedance, so pick the price that suits you.

03

Generating with Seedance 2.5

The steps are similar on OpenArt and Morphix Pro. The screenshots below are from Morphix Pro.

  • 1Go to the site you chose and open the Video section.
  • 2Add both reference images using the Add visual references option, the scale model first (Image 1) then the design (Image 2), since the prompt refers to them in that order.
  • 3Add the full video prompt from section one.
  • 4Choose the model (Seedance 2.5).
  • 5Choose the settings that suit you: dimensions (9:16), duration (10s), resolution, and audio (On).
  • 6Click Generate.
The images must be Reference

Add both images as "Reference", not as a "Start Frame". The shot starts on an empty site without the model, then the volumes are placed one by one, so the scale model image is where the shot arrives, not its first frame.

Save credits before generating

If you feel the video will cost too many credits, choose a low resolution like 480p in the settings, then Upscale the result as in the next section. This saves a lot of credits.

Video settings on Morphix Pro: both reference images with the prompt, Seedance 2.5, 9:16, 10s, 720p, audio On

Video settings on Morphix Pro: both reference images with the prompt, Seedance 2.5, 9:16, 10s, 720p, audio On

04

Upscale

The Upscale is done on the same site, and the engine differs from site to site. On Morphix Pro the engine is ByteDance; on OpenArt it is BytePlus vCube.

  • 1After generating the video, click on it inside the site to open its options menu.
The menu that opens after clicking the video, with the Upscale option

The menu that opens after clicking the video, with the Upscale option

  • 2Click Upscale, and the video opens with the Upscale settings.
  • 3Choose the settings that suit you, no need to copy mine. The cost is relatively low.
  • 4Click Generate.
My Upscale settings on Morphix Pro (ByteDance, 2k, 30 fps). Choose what suits you, you don't have to copy them

My Upscale settings on Morphix Pro (ByteDance, 2k, 30 fps). Choose what suits you, you don't have to copy them

05

The Free Alternative: Google Flow

  • 1Open Google Flow.
  • 2Add the prompt and both reference images (using the Ingredients option, not Frames, for the same reason mentioned above).
  • 3Choose the right settings and model (Omni 1.1 Flash), with 9:16 dimensions and a 10s duration.
  • 4Click Generate.
Choosing Ingredients, dimensions (9:16), the model (Omni 1.1 Flash), 720p, and duration (10s) on Google Flow

Choosing Ingredients, dimensions (9:16), the model (Omni 1.1 Flash), 720p, and duration (10s) on Google Flow

Final Result

One continuous 10-second take: a hand builds the wooden scale model floor by floor, then the camera accelerates backwards and, at the climax, the model turns into the real building in daylight, matching the design image.

Special Discount

If you'd like to subscribe to OpenArt or Morphix Pro, the links above give you a special discount when you sign up through them.

If you are still presenting your projects like it's 2019, this is your sign to stop. Click the link below to learn more.

I want to present better
Extra Resources

More Tools for Architects & Designers

If you're looking for more, my Gumroad store has ready-made libraries and prompts to help you present your projects better.