AI video

Runway for architecture: what Gen-4.5 and Aleph 2.0 do with renderings, and where buildings break

Runway for architecture explained: what Gen-4.5 and Aleph 2.0 do with renderings, where facades break, credit costs per plan and a six-image workflow.

9 min read

In this article

You have a set of renderings and want to know whether Runway is the right tool to make them move. Runway is one of the strongest general video tools available, but it is a clip tool: it turns one image into one short shot, or edits a clip you already have, and leaves the film to you. This guide covers which Runway models matter for architecture as of October 2026, what they do well with buildings, where facades and interiors break, what a six-clip project costs in credits, and when a one-step tool fits the job better.

Key takeaways

  • Runway's two models that matter most for architecture are Gen-4.5, which generates a clip from a rendering and a prompt, and Aleph 2.0, which edits existing footage up to 30 seconds at 1080p.
  • On Runway's pricing page Gen-4.5 costs 60 credits per 5 seconds, so the Standard plan's 625 monthly credits cover about ten five-second clips before any retries.
  • Runway's own research page lists object permanence as a known limitation of Gen-4.5, which on a building shows up as windows, mullions or railings that appear, vanish or change count.
  • Aleph 2.0 is most useful on footage whose geometry is already right, such as a walkthrough exported from your renderer or a drone clip of a finished project.
  • A six-image project film in Runway means six prompts, a review pass per clip, re-rolls, and an edit with licensed music that you assemble yourself.

Is Runway good for architecture?

Runway is good for architecture when you want control over individual shots or need to edit footage you already have. Gen-4.5 animates a single rendering from an image and a prompt, and Aleph 2.0 edits existing clips up to 30 seconds at 1080p, which suits a renderer walkthrough or a drone clip. It is not a one-step way to get a finished project film: you write a prompt per shot, check each clip for bending geometry, then edit the clips together and add music yourself.

That split decides most of what follows. If your job is "one hero clip for a pitch deck" or "make this daytime flythrough read as dusk," Runway is a strong choice. If your job is "a 30-second vertical Reel from six project images by Friday," the shot-by-shot work is the main cost, and it pays to price that in before you start.

Runway's current models, as of October 2026

Runway now hosts its own models alongside several third-party ones, all billed in the same credits. These are the entries that matter for renderings, taken from Runway's pricing page and its API model guide, as of October 2026.

Model What it does Input Web app credits API price
Gen-4.5 (Runway) Generates a new clip Text or an image 60 credits per 5 s USD 0.12 per second
Aleph 2.0 (Runway) Edits an existing clip Video 140 credits per 5 s USD 0.28 per second
Kling 4.0 (Kuaishou, hosted) Generates a clip, 3–30 s per the API guide Text, frames, references 118 credits per 5 s (89 with a discount through November 30) From USD 0.157 per second

The pricing page also lists ByteDance's Seedance 2.0 Pro at 160 credits per 4 seconds, and Seedance 2.5 among the Pro plan's models. Runway announced Gen-4.5 on December 1, 2025, describing "cinematic, photorealistic outputs with precise prompt adherence" (Runway Research); at launch it was text-to-video only, and the API guide now lists text and image input. Aleph 2.0 arrived on May 21, 2026, with a new interface called Edit Studio, and Runway says it is "available now on all paid Runway plans on our desktop web app" (Runway News).

Runway's help article on Gen-4.5 settings could not be loaded for this guide, so its exact durations and output resolution could not be verified as of October 2026. Check the generator's duration and resolution menus before you plan a 1080 × 1920 deliverable around it.

What Runway does well with renderings

Per-shot control

You decide which rendering becomes which clip and what the camera does in it. That matters for architecture, where one image might need a slow push-in toward the entrance and another a gentle lateral move along a facade. The prompt is your shot list, and each clip can be re-rolled until it behaves. For which moves keep verticals straight, see AI video prompts for architecture.

Editing footage whose geometry is already right

Aleph 2.0 is the more interesting model for many studios, because it starts from video rather than a single still. Runway describes "localized edits with precise input video preservation" and says the model "will apply your edit across relevant shots" in a multi-shot clip (Runway News). For architecture, that means a walkthrough exported from Enscape, Lumion or Twinmotion, where the camera path follows your real model, can be pushed toward dusk, rain or a different season without re-rendering. The same applies to a drone clip of a finished project with a parked truck you would rather not show. Any edit is still an AI interpretation, so compare the edited clip with the original frame by frame before it goes out.

Several models under one subscription

Because Kling 4.0 and Seedance models sit inside the same credit balance, you can run one rendering through two or three models and keep whichever holds the building best. That is a fairer test than reading anyone's ranking, including this one.

Where buildings break in Runway

Runway is unusually candid about Gen-4.5's limits. Its research page lists causal reasoning, where "effects sometimes precede causes," and object permanence, where "objects may disappear or appear unexpectedly across frames" (Runway Research). On people and sports, that reads as a quirk. On a building, it reads as a design change.

What breaks Where you see it How to reduce it
Window and mullion counts change Long facades, curtain walls Push in slowly; avoid long lateral moves
Verticals bow or lean Wide-angle exteriors, tall interiors Start from a two-point perspective image with straight verticals
Invented space past the frame edge Pans, pull-backs, orbits Keep the move inside what the image shows
Extra people or furniture appear Lobbies, living rooms Say what stays empty in the prompt, then review
Signage and lettering smear Storefronts, wayfinding Remove small text from the source image
Materials drift Stone, timber grain, terrazzo Shorter clips; reject and re-roll

The pattern behind all of them: an image-to-video model reads pixels and infers depth, but it never saw your plans. Anything outside the frame or hidden behind a column is a guess. The AI video quality checklist gives a frame-by-frame review routine that catches most of these before a client does.

What Runway costs for one project film

As of October 2026, Runway's pricing page lists a Free plan with 125 one-time credits and "a selection of generative creative models to test concepts," then three paid tiers. Gen-4.5 is named in the Standard plan's model list, and "no watermarks" is listed from Standard up (Runway pricing).

Plan Monthly price (annual billing) Credits per month Five-second Gen-4.5 clips, no retries
Standard USD 15 (USD 12) 625 About 10
Pro USD 35 (USD 28) 2,250 About 37
Max USD 95 (USD 76) 9,500 About 158

Now apply it to one project. Six five-second Gen-4.5 clips cost 360 credits. If half of them need one re-roll, which is normal on facades, that is 540 credits, close to the whole Standard allowance. Running the same six through Aleph 2.0 afterward to change the time of day would add 840 credits (six clips at 140 each), which only fits on Pro or above. Only the Max plan lists rollover ("roll over 1 month"), so plan around unused credits disappearing on the other tiers.

Those figures cover generation only. Music, captions and the edit are separate time, and the music needs a license that covers a business account posting its own work.

A six-image project workflow in Runway

If you decide Runway is the right tool, this routine keeps the project from sprawling.

  1. Pick six images that fill different roles. Context, approach, the main space, a quieter space, a material detail, a closing view. The six roles a project video needs explains why six angles of one facade make a weak film.
  2. Export at full size in the ratio you will publish. A 9:16 deliverable starts best from portrait or square sources; a wide image gets recomposed and edge details can drop out.
  3. Write one prompt template and change only the camera line. Keep the material, light and "no people, no added objects" lines identical across all six.
  4. Generate one clip per image and review it at full screen. Check window counts, verticals and the frame edges before you generate the next one.
  5. Re-roll or switch models for the clips that bend. Try the same image in Kling 4.0 inside your plan before rewriting the prompt a fifth time.
  6. Assemble the clips in your editor, add a licensed track and captions, and export a 1080 × 1920 file.
  7. Label it honestly. Unbuilt work is concept work; AI rendering disclosure wording has captions you can copy.

Runway or a one-step tool

The fair comparison is not "which is better" but "which job are you doing." Runway is a toolkit; a finished-film service makes one specific deliverable.

Runway A finished-film service such as Arch2Video
Input One image or one clip per generation Exactly six images from one project
Output Individual clips you edit together A 30-second 1080 × 1920 MP4 of six connected five-second scenes
Prompting A prompt per clip None; an AI director plans the scenes and their order
Editing existing footage Yes, with Aleph 2.0 No
Music You license and add it Original soundtrack included (Still Air by default, Quiet Piano, Concrete, or none)
People Avoidable through prompting, not guaranteed Always a small cast of fictional adults in at least one scene
Price, as of October 2026 Monthly plan from USD 15, credits per clip USD 15 per video, one-time, no subscription

Choose Runway when you need a specific shot, a non-vertical format, or edits to footage you already trust. Choose Arch2Video when the deliverable is a 30-second vertical project film and you would rather review one finished file than direct six clips; generation usually takes an hour or more and you are emailed when it is ready. The comparison of AI video generators for architects sets Runway beside Veo, Kling and Luma if you are still choosing.

Frequently asked questions

Can Runway make a walkthrough from my Revit or SketchUp model?

No. Runway works from images and video, not from a 3D model, so it cannot follow a measured route through your plan. For a true walkthrough, export the camera path from your renderer, then use Aleph 2.0 if you want to change the light or weather on that footage. Your renderer's own video export is the measured version; Runway is the interpretive one.

Is Runway's free plan enough to test my renderings?

It is enough to get a feel for the interface, not to judge Gen-4.5 on your project. As of October 2026 the Free plan gives 125 one-time credits and "a selection" of models, Gen-4.5 is named under the Standard plan, and "no watermarks" is listed from Standard up. A one-month Standard plan is the cheaper honest test: six images, one clip each, one re-roll for the worst two.

Why does my building change shape halfway through a Runway clip?

Because the model is predicting what the building looks like from the next viewpoint without any geometry to check against. Runway itself lists object permanence among Gen-4.5's limitations. Shorter clips, slower camera moves that stay inside the frame, and source images with straight verticals and no small text all reduce the drift, but none removes it; review every clip before you use it.

Should I use Runway's own model or Kling inside Runway for architecture?

Test both on the same rendering, because results vary by image more than by reputation. As of October 2026, Gen-4.5 costs 60 credits per five seconds and Kling 4.0 costs 118 (89 during Runway's discount through November 30), so a side-by-side test of one image costs under 200 credits. Keep whichever holds the window pattern and verticals, and use the same model for all six clips so the film looks consistent. Kling AI for architecture covers Kling's own app and keyframe controls.