The most complete, up-to-date comparison of AI video generation models — which model, via which API, at what price, and how fast.
Unlike other lists that just dump links, this one answers the question developers actually have: "I need to generate video — which model do I pick, and where do I call it?" Every model below is mapped to the APIs that serve it, with indicative pricing, speed, and max resolution/duration.
💡 Prices are per second of output video (retail API rates verified July 2026) and move fast — always confirm against the provider. Latency is a rough "time to first result" at default settings.
- Open-Generative-AI — curated hub of open generative-media tools and pipelines
- Text-To-Video-AI — generate a finished video from a text prompt end-to-end
- Open-AI-Micro-Drama-Generator — multi-scene AI micro-drama pipeline
- Veo-4-API — Python wrapper for Google Veo 4
- Seedance-2-API — Python wrapper for ByteDance Seedance 2
- awesome-seedance-2.5-api-prompts — prompt library for Seedance 2.5
- ai-creator-academy — free curriculum teaching creators how to monetize the models compared in this list
- Flux-3-Dev-API — Python wrapper for Black Forest Labs' FLUX 3 (Dev variant) — text-to-image, image-to-image, text-to-video, image-to-video
- awesome-flux-3-api-prompts — FLUX 3 API guide, prompts, and parameters
- Commercial models (closed, API-only)
- Open-source models (self-host or API)
- Task-specific: image-to-video
- Avatar & talking-head models
- Real-time & interactive video
- Video upscaling & enhancement
- Benchmarks & leaderboards
- How to choose
- Where to run them (API providers)
- Contributing
| Model | Maker | Best for | APIs | Price / sec | Max res / length | Notes |
|---|---|---|---|---|---|---|
| Veo 3.1 | Realism + native audio | Gemini API, Fal, Replicate, MuAPI | $0.15 Fast · $0.40 Quality (Lite $0.03–0.05) | 1080p/4K / ~8s | Native audio; 4K at premium | |
| Sora 2 | OpenAI | Coherence, long shots | OpenAI API | $0.10 Std · $0.30–0.50 Pro | 720p–1080p / ≤25s | |
| Kling 3.0 | Kuaishou | Motion, prompt adherence | Fal, Replicate, MuAPI | ~$0.10 (up to $0.20 w/ audio); ~$0.075 via resellers | 1080p / 5–10s | Motion Control variant available |
| Runway Gen-4.5 | Runway | Creative control, editing | Runway API, Fal | $0.12 (Turbo $0.05) | 1080p / ~10s | Strong editing suite |
| Seedance 2.0 | ByteDance | Value, speed, quality | Fal, Replicate, MuAPI | ~$0.03 Fast · ~$0.05 Pro | 1080p / 5–10s | 2.5 in beta: 30s clips, 50 refs, pricing TBA |
| Hailuo 2.3 | MiniMax | Character motion | Fal, Replicate, MuAPI | $0.045 (768p) · ~$0.017 (512p) | 1080p / ≤10s | 30–90s generation |
| Luma Ray 3 | Luma | Fast iteration, HDR | Luma API, Fal | ~$0.21 (HDR ~2×) | 1080p+HDR / 5–10s | Per-second billing |
| Pika 2.2 | Pika | Effects, stylization | Pika API | ~$0.05 | 1080p / ~5s | Cheapest closed model |
| Model | Maker | License | Best for | APIs | Self-host VRAM |
|---|---|---|---|---|---|
| Wan 2.2 | Alibaba | Apache-2.0 | Quality leader; T2V+I2V+edit in one | Fal, Replicate | ~24GB+ |
| HunyuanVideo 1.5 | Tencent | Custom OSS | Natural motion & physics | Fal, Replicate | ~40GB+ |
| LTX-2.3 | Lightricks | OpenRAIL | Fast; native 4K + synced audio | Fal | ~12GB+ |
| CogVideoX | Zhipu / THUDM | Apache-2.0 | Research, fine-tuning | self-host | ~18GB+ |
| Mochi 1 | Genmo | Apache-2.0 | High-fidelity motion | Fal, self-host | ~24GB+ |
| Open-Sora 2.0 | HPC-AI Tech | Apache-2.0 | Fully open pipeline + weights | self-host | ~24GB+ |
| SkyReels-V3 | Skywork | Custom OSS | Cinematic / film-style shots | self-host | ~24GB+ |
| NVIDIA Cosmos | NVIDIA | NVIDIA OpenModel | World models, robotics/sim | self-host | ~40GB+ |
| AnimateDiff | Community | Apache-2.0 | Add motion to SD image models | self-host | ~10GB+ |
Most models above also do image-to-video (I2V). Strongest I2V specifically:
- Kling 3.0 — best all-round I2V motion
- Seedance 2.0 — best price/quality for I2V
- Wan 2.2 — best open-source I2V
- Luma Ray 3 — fastest I2V for iteration
Portrait/presenter video from a photo or actor + script. These bill by subscription or per-minute credits (not per-second like the raw generators), so the pricing column names the model, not a rate.
| Tool | Maker | Best for | API | Pricing model |
|---|---|---|---|---|
| HeyGen | HeyGen | Studio avatars, dubbing | Yes (+ streaming SDK) | Credits / subscription |
| Synthesia | Synthesia | Enterprise training videos | Yes | Seat / subscription |
| D-ID | D-ID | Real-time talking portraits | Yes (streaming) | Per-minute credits |
| Hedra | Hedra | Expressive character audio→video | Yes | Credits |
| Tavus | Tavus | Conversational video agents | Yes | Per-minute |
| Captions (Mirage) | Captions | Fully generated AI actors | Yes | Credits / subscription |
| Hour One | Hour One | Scalable presenter video | Yes | Subscription |
Open-source alternatives: OmniHuman-1, LivePortrait, EchoMimicV2, MuseTalk, OmniAvatar (self-host, see repos).
Emerging category — sub-second/streaming generation for games, live avatars, and interactive worlds.
- Decart (Lucy) — real-time world/video generation
- Krea Realtime 14B — open-weights real-time generation
- LongLive — long-duration real-time generation
- PixVerse — fast interactive generation
- Hunyuan-GameCraft / world models — playable generated environments
Post-process generated (or real) footage — upscale, interpolate, denoise.
| Tool | Type | Best for |
|---|---|---|
| Topaz Video AI | Commercial | Highest-quality upscale + interpolation |
| FlashVSR | Open source | Fast video super-resolution |
| Video2X | Open source | Free upscaling (waifu2x/Real-ESRGAN) |
| REAL Video Enhancer | Open source | Interpolation + upscaling GUI |
Don't trust a maker's own demo reel — check independent evals before committing:
- VBench / VBench-2.0 — 16-dimension automated quality benchmark
- Artificial Analysis Video Arena — Elo-style human-preference leaderboard across commercial + open models
- Video-Bench — human-aligned evaluation suite
- Best realism / audio → Veo 3.1 (Sora 2 works too, but its API sunsets Sept 2026)
- Best value at scale → Seedance 2.0 (
$0.03/s) or Pika 2.2 ($0.05/s) - Fastest iteration → Luma Ray 3 or LTX-2.3
- Fully open / self-hosted → Wan 2.2 (quality) or LTX-2.3 (speed)
- Editing & creative control → Runway Gen-4.5
Aggregators that expose many of the above behind one API/key:
- MuAPI — unified API across image + video models (Kling, Veo, Seedance, Hailuo, Wan, and more), one key, one billing
- Fal — fast inference, broad model catalog
- Replicate — pay-per-run, large community model catalog
Native APIs (single-vendor): Google Gemini (Veo), OpenAI (Sora), Runway, Luma, Pika, MiniMax.
PRs welcome. When adding a model, keep the table columns filled — a row without provider + price + speed isn't useful. New models go in the correct table (commercial vs open-source) and stay sorted by relevance.
Maintained alongside Open-Generative-AI. Found it useful? ⭐ the repo.