The best AI video generator, tested
Every AI video model on this site received the same nine prompts, requesting five seconds where supported: a product turntable, a person speaking, a title card, a camera move, a spill, hands at work, a market, a woodblock print and a still photo brought to life. We paid for every clip, recorded the wait and the cost, and kept the failures. This page ranks them prompt by prompt, with the whole roster visible under each verdict. The September 18 additions are H3 Max, H3 Max Turbo, LTX 2.3 Pro and PixVerse V6; FLUX 3 Max joined on October 19, and on October 22 LTX 2.5 Pro and Fast, Gemini Omni Flash 1.1, Wan 3.0 Prime, Seedance 2.0 Mini, Kandinsky 6.0 Pro and Lite, Grok Imagine Video 1.5 Lite and HeyGen Video. Kling 4.0 joined on October 23. New observations use sampled frames across the clips; audio has not been scored by ear. Longer workflow tests live on the relevant model pages and are excluded from this short-shot ranking.
Anvisha Pai, Co-founder & CEO, Voyager
The short answer
- 1.Veo 3.1 Fast is the model most people should ship: clean text, the best crowd and camera work, native audio, under two minutes a clip. $0.15 per second with audio, 720p or 1080p via fal.
- 2.FLUX 3 Max is the new default for cinematic and realistic shots: the best spill and a dense, legible market, every clip back in about 16 seconds, at $0.17 a second at 720p (a quarter of that until November 7). $0.17 per second at 720p via fal.
- 3.Seedance 2.0 is the best-looking set on the roster and the most natural talking head, at five times Kling's price. $0.30 per second at 720p, 16:9 via fal.
- 4.Wan 3.0 is the best value: steady, consistent and 30-second clips, with a content checker that refused one prompt. $0.10 per second at 720p via fal.
- 5.Kling 3.0 Standard for 15-second clips and multi-shot storyboards without on-screen text, if you can wait for fal's queue. $0.13 per second, audio on via fal.
- 6.Veo 3.1 when a hero shot needs the last bit of polish or 4K; otherwise Fast is the same look for a quarter of the price. $0.40 per second with audio, 720p or 1080p via fal.
- 7.H3 Max Turbo is a low-cost draft option with readable title text in this run, at $0.04 a second at 768p. $0.04 per second at 768p via fal.
- 8.HeyGen Video is the cheapest start with sound, added on 22 October: $0.10 a five-second 768p clip, a correct title card and a dense market, though it cuts mid-shot when it likes. $0.02 per second at 768p via fal.
Every model, measured
| Model | Vendor | Price | Seconds to make one clip | 9 clips cost |
|---|---|---|---|---|
| FLUX 3 Max | fal / Black Forest Labs | $0.17 per second at 720p via fal | 16 to 39 | $1.91 |
| Gemini Omni Flash 1.1 | Google DeepMind | $0.10 per second at 720p via fal | 31 to 60 | $4.50 |
| Gemini Omni Flash | Google DeepMind | $0.10 per second of 720p video via Gemini API | 36 to 64 | $5.85 |
| Grok Imagine Video 1.5 | xAI | $0.08 per second via xAI API | 36 to 49 | $6.30 |
| Grok Imagine Video 1.5 Lite | xAI | $0.03 per second at 720p via fal | 21 to 41 | $1.36 |
| HeyGen Video | HeyGen / MiniMax | $0.02 per second at 768p via fal | 21 to 41 | $0.90 |
| H3 Max Turbo | fal / MiniMax | $0.04 per second at 768p via fal | 6 to 6 | $0.90 |
| H3 Max | fal / MiniMax | $0.08 per second at 768p via fal | 6 to 26 | $1.80 |
| MiniMax H3 | MiniMax | $0.06 per second at 768p via fal | 52 to 182 | $2.70 |
| Hailuo 2.3 | MiniMax | $0.28 per six-second clip, standard 768p via fal | 98 to 197 | $2.52 |
| Kandinsky 6.0 Pro | Kandinsky Lab | $1.35 per five-second 480p clip, text-to-video via fal | 830 to 1553 | $12.20 |
| Kandinsky 6.0 Lite | Kandinsky Lab | $0.16 per five-second 480p clip, text-to-video via fal | 193 to 777 | $1.45 |
| Kling 4.0 | Kuaishou | $0.16 per second at 720p via fal | 145 to 224 | $7.07 |
| Kling 3.0 Standard | Kuaishou | $0.13 per second, audio on via fal | 129 to 729 | $5.67 |
| Kling 3.0 Pro | Kuaishou | $0.17 per second, audio on via fal | 134 to 692 | $7.56 |
| LTX 2.5 Pro | Lightricks | $0.12 per second at 720p via fal | 47 to 143 | $6.48 |
| LTX 2.5 Fast | Lightricks | $0.09 per second at 720p via fal | 47 to 137 | $4.86 |
| LTX 2.3 | Lightricks | $0.08 per second at 1080p via fal | 54 to 73 | $4.32 |
| Luma Ray 3.2 | Luma AI | $0.30 per five-second sdr clip at 720p via fal image-to-video | 62 to 187 | $9.00 |
| PixVerse V6 | PixVerse | $0.06 per second at 720p with audio via fal | 49 to 88 | $2.70 |
| Seedance 2.0 | ByteDance | $0.30 per second at 720p, 16:9 via fal | 88 to 148 | $13.61 |
| Seedance 2.5 | ByteDance | $0.46 per second at 720p, 16:9 via fal | 170 to 1578 | $18.49 |
| Seedance 2.0 Mini | ByteDance | $0.007 per 1,000 tokens via fal | 147 to 458 | $6.80 |
| Sora 2 | OpenAI | $0.10 per second at 720p, sora-2 via OpenAI API | 67 to 200 | $3.60 |
| Veo 3.1 Fast | Google DeepMind | $0.15 per second with audio, 720p or 1080p via fal | 52 to 96 | $8.10 |
| Veo 3.1 | Google DeepMind | $0.40 per second with audio, 720p or 1080p via fal | 67 to 108 | $21.60 |
| Veo 3.1 Lite | Google DeepMind | $0.05 per second with audio, 720p via fal | 37 to 102 | $2.70 |
| Wan 3.0 | Alibaba | $0.10 per second at 720p via fal | 128 to 194 | $4.00 |
| Wan 3.0 Prime | Alibaba | $0.14 per second at 720p via fal | 84 to 103 | $5.60 |
| Wan 2.7 | Alibaba | $0.10 per second at 720p via fal | 41 to 134 | $4.00 |
Seconds and cost are from our own runs of the same 9 prompts through every model, one try each.
Person speaking
This prompt in full →Seedance 2.0 gives the most natural face and delivery; Veo 3.1 Fast and Wan 3.0 are close for much less. Grok burned a subtitle in, Sora 2 barely moved, Hailuo 2.3 is silent. Luma Ray 3.2 is silent but its performance is natural. LTX 2.5 Pro, added on 22 October, is the liveliest of that batch.
Animated title card
This prompt in full →Veo in both main tiers, Seedance 2.0, MiniMax H3 and Hailuo 2.3 spelled the title. Both Kling tiers garbled it; both Wan endpoints refused it. H3 Max, Turbo, PixVerse V6 and FLUX 3 Max resolve into readable titles; LTX 2.3 garbles the lettering. Of the 22 October additions, Seedance 2.0 Mini set it exactly as asked; Grok Lite, Gemini Omni Flash 1.1 and HeyGen Video spelled it; both LTX 2.5 tiers and Kandinsky 6.0 Pro garbled it, Wan 3.0 Prime refused it, and Kandinsky 6.0 Lite showed a fox instead. Kling 4.0, added on 23 October, is the first Kling to spell it, both lines and the date.
Product turntable
This prompt in full →Veo 3.1 Fast, Hailuo 2.3, Wan 3.0 and Kling Standard held the product through the turn. Wan 2.7 and Gemini added cuts; Seedance 2.5 opened on a blur. Of the 22 October additions only Wan 3.0 Prime kept the dripper's shape; most of the rest added a handle or changed the vessel. Kling 4.0, added on 23 October, kept the shape but cut twice on the one-shot prompt.
Image to video
This prompt in full →Wan 3.0, Seedance 2.0 and Kling kept the photographed product exactly. Gemini gave it a handle, Seedance 2.5 refused the job. FLUX 3 Max kept the product but filled it like a cup. Of the 22 October additions only Kandinsky 6.0 Pro kept the product and poured into it; most filled it like a cup, and Gemini Omni Flash 1.1 and Seedance 2.0 Mini grew a handle.
Camera move
This prompt in full →Veo 3.1 Fast's aerial is the most convincing; Wan 3.0 and Gemini Omni Flash move the camera with the town intact. LTX 2.5 Pro, added on 22 October, flies a convincing push into the harbour.
Physics
This prompt in full →FLUX 3 Max throws the most convincing spill, the water pooling and the cubes sliding apart; Veo 3.1 is close and Veo 3.1 Lite has the biggest splash. Wan 2.7 grew the ice cubes; Seedance 2.5 turned the water to haze. LTX 2.3 suspends the glass in its middle sampled frame; PixVerse rounds the ice and loses the glass late in the shot. No 22 October addition beats FLUX 3 Max; several multiplied the ice cubes.
Hands at work
This prompt in full →Kling 3.0 Pro poured best with the props intact. Half the roster swapped the filter for a cup. Wan 3.0 Prime, Gemini Omni Flash 1.1, LTX 2.5 Pro and Seedance 2.0 Mini, added on 22 October, all kept the props.
Busy scene
This prompt in full →Veo 3.1 Fast's market is the densest and most legible; FLUX 3 Max is as busy, with the dog and the busker both present; Wan 3.0, Seedance 2.0 and MiniMax H3 hold up. Kling's people look painted. Gemini Omni Flash 1.1, Wan 3.0 Prime and HeyGen Video, added on 22 October, hold up. Kling 4.0, added on 23 October, drops the painted look: dense and natural, with the busker and the dog on its lead.
Specific art style
This prompt in full →Veo 3.1 stays a woodblock print in every frame; MiniMax H3 and Veo Lite hold the style with lively waves. Hailuo washed out. Wan 3.0 Prime, added on 22 October, is the truest print of its batch.
Head to head
- FLUX 3 Max vs Seedance 2.5
- Gemini Omni Flash 1.1 vs Gemini Omni Flash
- Grok Imagine Video 1.5 Lite vs Grok Imagine Video 1.5
- H3 Max vs H3 Max Turbo
- Hailuo H3 Max vs Kling 3.0
- HeyGen Video vs H3 Max
- Kandinsky 6.0 Pro vs Kandinsky 6.0 Lite
- Kling 4.0 vs Kling 3.0 Pro
- Kling 3.0 vs Veo 3.1 Fast
- LTX 2.5 Pro vs LTX 2.5 Fast
- Hailuo (MiniMax H3) vs Kling 3.0
- Seedance 2.0 Mini vs Seedance 2.0
- Seedance 2.0 vs Kling 3.0
- Seedance 2.0 vs Veo 3.1
- Seedance 2.5 vs Kling 3.0
- Seedance 2.5 vs Veo 3.1
- Sora 2 vs Veo 3.1
- Wan 3.0 Prime vs Wan 3.0
- Wan 3.0 vs Kling 3.0
Frequently asked questions
What is the best AI video generator in 2026?
On the same nine clips, Veo 3.1 Fast is the best all-round model for the money, Seedance 2.0 looks best, Wan 3.0 is the best value, and Kling 3.0 wins on clip length and storyboards but cannot render text; Kling 4.0, added on 23 October, can. Every verdict on this page links to the clips behind it.
What is the best free AI video generator?
None of these models has a free API. The Kling app gives 66 credits a day with a watermark, Dreamina and Hailuo give daily credits, fal's Wan 3 page offers five free 720p clips a day, and Google's Gemini app includes a few Veo clips on paid plans. Each model page lists the free tier in detail.
Which AI video generator has audio?
All of them here except Hailuo 2.3: Veo 3.1, Kling 3.0, Seedance, Sora 2, Grok Imagine Video, Wan 2.7 and 3.0, MiniMax H3, FLUX 3 Max and Gemini Omni Flash all generate sound with the picture. So do the October 22 additions: LTX 2.5, Gemini Omni Flash 1.1, Wan 3.0 Prime, Seedance 2.0 Mini, Kandinsky 6.0, Grok Imagine Video 1.5 Lite and HeyGen Video. Kling 4.0, added on October 23, makes stereo sound on every clip, with no switch to turn it off.
How much does AI video cost?
From $0.28 for a silent six-second Hailuo clip to $2.40 for six seconds of Veo 3.1 with audio, at 720p, on fal, read on 10 September 2026. The table above shows what our nine clips cost on every model. HeyGen Video, added on 22 October, is cheaper still: $0.10 for five seconds at 768p with sound.
Is Sora 2 still available?
No. OpenAI shut the API down on 24 September 2026, after closing the app in April. Its clips, made before the shutdown, stay on this page for comparison, and its own page names the replacements.
Voyager
Run supported models inside Voyager
Voyager is an agent for creative work that runs supported models. Brief it, and it picks the model for the job, makes the thing and hands back something you can edit. Create an account to get started.