Spec Guide
Frames per second in AI video generation
What FPS means, how it changes your output, and which models give you the most
What is FPS?
Frames per second is how many individual images the model generates for each second of video. More frames means smoother motion. Fewer frames means choppier output — sometimes stylistic, usually just a hardware limitation.
The number matters because it determines whether your output looks cinematic, hyperreal, or like a flipbook. Most people have seen the difference without knowing the term: film looks different from a soap opera, and that's largely an FPS difference. Film is shot at 24fps. Soap operas and live sports are broadcast at 30fps or higher. Your brain reads the difference as “prestige” vs “immediate” even when it can't articulate why.
In AI video generation, FPS is set by the model — you don't always get to choose. Some models lock you to 24fps. Others offer a range. A few push to 50–60fps for specialized use cases. Knowing what each model supports before you start saves a lot of wasted generation time.
How FPS changes what you see
Noticeable frame-to-frame jumps. Only useful for very stylized or experimental work.
CogVideoX-5B
The cinema standard. Films are shot at 24fps. Natural for narrative content, cinematic B-roll, and anything that should feel "filmy."
Wan 2.1, Wan 2.2, Wan 2.6, HunyuanVideo, SkyReels V2, Seedance 2.0, Luma Ray 3, Pika 2.5, Grok Imagine, Adobe Firefly
Slightly smoother than 24fps. Broadcast and gaming standard. Used for social media, YouTube, and product demos — content that needs to feel present rather than cinematic.
Mochi 1, Kling O1, Runway Gen-4.5, Veo 3.1, Sora 2, Hailuo 2.3
Sports, VR, product spins, macro detail. Can look too smooth for narrative (the "soap opera effect"). Useful for specific production needs, not everything.
LTX-2.3 (up to 50fps), LTX-2 (up to 50fps), Kling 3.0 (up to 60fps)
All models ranked by FPS
24 models, sorted highest to lowest. Click any model to view its full spec sheet.
| Source | Cost/sec | On Floyo | |||
|---|---|---|---|---|---|
| Kling 3.0 Kuaishou | Up to 60 48–60 fps | Native 4K (3840x2160) at up to 60fps | Closed | $0.10 | — |
| LTX-2.3 Lightricks | Up to 50 48–60 fps | 4K | Open | $0.04 | Yes |
| LTX-2 Lightricks | Up to 50 48–60 fps | 4K | Open | $0.04 | Yes |
| Mochi 1 Genmo | 30 30 fps | 480p (720p coming) | Open | $0.33/clip | — |
| Runway Gen-4.5 Runway | 24-30 30 fps | 1080p (upscaled 4K) | Closed | ~$0.15 (credits) | — |
| Veo 3.1 Google DeepMind | 24-30 30 fps | 1080p and 4K | Closed | $0.20 | Yes |
| Sora 2 OpenAI | 24-30 30 fps | 1080p | Closed | $0.15 | Yes |
| Hailuo 2.3 MiniMax | 24-30 30 fps | 1080p | Closed | ~$0.25/clip | — |
| Kling O1 Kuaishou | 30 30 fps | 4K | Closed | TBD | — |
| Wan 2.2 Alibaba (Tongyi Lab) | 24 24 fps | 720p | Open | Self-host | Yes |
| HunyuanVideo 1.5 Tencent | 24 24 fps | 720p | Open | $0.06 | Yes |
| SkyReels V2 Skywork AI | 24 24 fps | 720p | Open | Self-host | — |
| Wan 2.6 Alibaba (Tongyi Lab) | 24 24 fps | 720p / 1080p | Closed | $0.05 | Yes |
| Seedance 2.0 ByteDance | 24 24 fps | 1080p | Closed | $0.14 | — |
| Luma Ray 3 Luma AI | 24 24 fps | 4K (HDR EXR) | Closed | ~$0.50-1.00 | — |
| Pika 2.5 Pika Labs | 24 24 fps | 1080p | Closed | ~$0.15 | — |
| Vidu 2.0 Shengshu Tech | 24 24 fps | 1080p | Closed | ~$0.04 | — |
| PixVerse 5.5 PixVerse | 24 24 fps | 1080p | Closed | TBD | — |
| Adobe Firefly Video Adobe | 24 24 fps | 4K | Closed | Credits | — |
| Open-Sora 2.0 HPC-AI Tech | 24 24 fps | 720p | Open | Self-host | — |
| Wan 2.1 Alibaba (Tongyi Lab) | 24 24 fps | 720p | Open | Self-host | Yes |
| Grok Imagine xAI | 24 24 fps | 720p | Closed | $0.05/sec | Yes |
| CogVideoX-5B Tsinghua / Zhipu AI | 8 Sub-24 fps | 720p (1360x768) | Open | Self-host | — |
| Higgsfield Studio Higgsfield.ai | Varies Sub-24 fps | Varies by model | Closed | Subscription | — |
Which FPS should you pick?
Making a short film or narrative content?
24 fpsIt's the standard for a reason — film looks like film at 24fps. The slight motion blur between frames is what makes cinema feel cinematic.
Wan 2.2, SkyReels V2, Luma Ray 3
Compare these models →Making social media content or product demos?
30 fpsSlightly smoother than cinema without looking hyperreal. Better for scroll-stopping content that needs to feel present and immediate.
Runway Gen-4.5, Veo 3.1, Kling 3.0
Compare these models →Making product spins, macro shots, or VR content?
48–60 fpsMaximum smoothness for content where fluidity matters more than cinematic feel. Every frame counts when showing fine detail in motion.
LTX-2.3, Kling 3.0
Compare these models →Run high-FPS workflows on Floyo
Browser-based ComfyUI. No setup, no GPU required.
LTX 2.3 Pro Image to Video
Upload a still image and describe the motion you want. The model reads composition, lighting, and depth from your image, then animates it with prompt-controlled camera moves, particle effects, and environmental dynamics. Supports optional end-frame for locked start/finish transitions. Up to 2160p with built-in audio generation.
LTX 2.3 Audio to Video
Feed in an audio file and the model generates video that follows the rhythm, intensity, and structure of the sound. Works with music, speech, or sound effects. Fully automated pipeline with no manual parameter tuning required. Ideal for music visuals, audio-reactive content, and quick audio-driven animations.
LTX 2.3 Pro Text to Video
Generate video from a text prompt using the Pro flow. Higher fidelity output with enhanced detail and stability across longer sequences. Supports resolutions up to 4K, multiple FPS options (24/25/48/50), and durations up to 20 seconds. Built-in audio generation included.
LTX 2.3 T2V (Community)
Community-built text-to-video workflow using LTX 2.3. Lightweight setup for quick text prompt to video generation.
LTX 2 19B Fast Text to Video
Text-to-video generation using LTX 2's 19B Fast checkpoint. Optimized for speed over maximum quality, suited for rapid iteration and prototyping. Open source model running on Floyo's cloud infrastructure.
Compare FPS alongside resolution, audio, cost, and more
Select any models from the browse view to run a full side-by-side comparison.
Explore all models →