Runway Gen-4.5 is the studio's frontier foundation model, delivering photorealistic, physics-aware video with the prompt precision that took it to the top of the Artificial Analysis leaderboard.
Scored 1,247 Elo on the Artificial Analysis Text-to-Video leaderboard at launch, edging out Google Veo 3, Kling 2.5 and OpenAI Sora 2 Pro.
A stronger world model gives motion real weight and momentum, with cause-and-effect and human movement that hold up shot to shot.
Six aspect ratios from vertical 9:16 to ultrawide 21:9, plus Image-to-Video control, with more modes such as Keyframes and Video-to-Video announced by Runway.
Built on NVIDIA Hopper and Blackwell GPUs, it keeps the responsiveness of Gen-4 while stepping up fidelity.
Runway Gen-4.5 is Runway's frontier video generation model, released on December 1, 2025 as the successor to the Gen-4 family. Runway positions it as a general-purpose foundation model for world modeling, built on gains in pre-training data efficiency and post-training technique rather than raw scale alone. At launch it took the top spot on the Artificial Analysis Text-to-Video leaderboard with 1,247 Elo, ahead of Google Veo 3, Kling 2.5 and OpenAI Sora 2 Pro.
The model's headline strength is a stronger internal world model: motion carries realistic weight and momentum, characters move and emote convincingly, and complex scenes stay coherent while closely following the prompt. Gen-4.5 debuted as text-to-video; Runway added Image-to-Video in January 2026 and has said more control modes, such as Keyframes and Video-to-Video, are on the way. As of August 2026 it remains Runway's frontier model, with no Gen-5 announced.
Gen-4.5 scored 1,247 Elo on the Artificial Analysis Text-to-Video leaderboard at launch, placing first ahead of Google Veo 3 (1,226), Kling 2.5 (1,225) and OpenAI Sora 2 Pro (1,206) in blind human preference testing.
Improved world modeling gives objects and characters believable weight, momentum and interaction, so motion and cause-and-effect look natural rather than drifting or sliding.
The model closely interprets detailed prompts — subject, action, camera and style — producing complex, well-composed scenes that match what you described.
Generates photorealistic and stylized characters with lifelike facial detail and expression, holding a consistent look across a shot.
Supports 16:9, 9:16, 1:1, 4:3, 3:4 and ultrawide 21:9, covering social verticals, square posts and widescreen cinematic framing from one model.
Beyond text-to-video, Gen-4.5 supports Image-to-Video (added January 2026), letting you drive a shot from a real, generated or illustrated still; Runway says more control modes such as Keyframes and Video-to-Video are on the way.
Running on NVIDIA Hopper and Blackwell GPUs, Gen-4.5 preserves the speed and efficiency of Gen-4 while raising output fidelity.
| Primary modality | Text-to-video |
|---|---|
| Control modes | Image-to-Video (Keyframes and Video-to-Video announced, not yet available) |
| Clip durations | 2–10 s (5 s or 10 s on SharkFoto) |
| Max resolution | 720p (4K only via Runway's separate Upscale tool) |
| Aspect ratios | 16:9, 9:16, 1:1, 4:3, 3:4, 21:9 |
|---|---|
| Orientation | Landscape, portrait, square, ultrawide |
| Launch benchmark | 1,247 Elo, #1 on Artificial Analysis Text-to-Video |
|---|---|
| Compute | NVIDIA Hopper & Blackwell GPUs |
| Access | Runway platform, API and select partners |
Generate establishing shots, dynamic sequences and moody cinematic moments in 21:9 or 16:9 with realistic motion and lighting.
Produce vertical 9:16 and square 1:1 clips for Reels, TikTok, Shorts and Stories directly in the right frame.
Create polished product shots and ad concepts with photorealistic detail and precise prompt control for fast creative iteration.
Animate expressive, consistent characters — photorealistic or stylized — for narrative pieces, explainers and concept films.
Bring a real, generated, sketched or illustrated still to life by driving motion from a first-frame image.
Gen-4.5 keeps Gen-4's speed and six aspect ratios while improving physics, prompt adherence and overall quality, which took it to #1 on the Artificial Analysis leaderboard at launch.
| Feature | Runway Gen-4 | Runway Gen-4.5NEW |
|---|---|---|
| Artificial Analysis rank | Below frontier leaders | #1 Text-to-Video at launch, 1,247 Elo |
| World model / physics | Strong consistency | Improved weight, momentum and cause-and-effect |
| Prompt adherence | Good | Higher fidelity to complex prompts |
| Aspect ratios | Six ratios incl. 21:9 | Same six ratios (16:9, 9:16, 4:3, 3:4, 1:1, 21:9) |
| Speed | Fast, efficient | Retains Gen-4 speed at higher quality |
The model can still misorder cause and effect, so an effect occasionally appears before the action that should produce it.
Objects may occasionally disappear, reappear or shift unexpectedly across frames in complex or fast-moving scenes.
Generated actions tend to resolve favorably by default, which can make outcomes look unrealistically successful without careful prompting.
Individual generations are capped at short durations (up to 10 seconds; 5 s or 10 s on SharkFoto), so longer sequences require stitching multiple clips together.
Turn a prompt or a single image into a cinematic, physics-aware 5- or 10-second clip in any of six aspect ratios, right on SharkFoto.
Try Runway Gen-4.5 now