What are the best AI video models: Seedance, Kling, Veo and more?

FLORA runs the main AI video models from one account, including Seedance 2.5, Kling 3.0 Pro, Veo 3.1, MiniMax H3 Max, Gemini Omni 1.1 Flash, FLUX 3 and WAN 3.0. Quotes for one image-to-video clip at each model's default settings ranged from $0.48 to $3.84. Our 4-second Seedance 2.5 clip cost $1.996.

At a glance

The options

1. FLORA

   The canvas and MCP server that run every model below from one account. Each clip is quoted in dollars before it runs and saved on the canvas next to the still it came from.

   Best for: Teams that want to use the right video model for each shot without a separate account for each

   Pros

   • 299 video model entries on one account, including Seedance 2.5, Kling 3.0, Veo 3.1, MiniMax H3 and WAN 3.0.
   • One free quote call priced the same clip on eight setups, from $0.48 to $3.84, before we spent anything.
   • Stills, clips and audio on one canvas, driven from Claude, Cursor, Codex or Claude Code.

   Cons

   • Clips take minutes: our Seedance 2.5 run took 3 minutes 54 seconds, so let the agent poll while you work.

2. Seedance 2.5

   Ids include t2v-gengateway-seedance-2-5-t2v and i2v-gengateway-seedance-2-5-i2v. 4 to 30 seconds at 480p, 720p or 1080p, aspect ratios from 9:16 to 21:9 or auto. Quoted $2.483 for 5 seconds and $1.987 for 4 seconds at 720p. Outputs carry an audio track.

   Best for: Long single takes with sound, built from approved stills and references

   Pros

   • Up to 30 seconds in one pass, with audio generated together with the picture (ByteDance, July 2026).
   • Takes heavy reference input: ByteDance lists up to 30 images, 10 videos and 10 audio clips.
   • Among the top three for both text-to-video and image-to-video on the Artificial Analysis leaderboards in October 2026.
   • In FLORA: seven input shapes, a dollar quote first, and the clip saved next to the still it came from.

   Cons

   • ByteDance says complex motion physics and scenes with several interacting subjects still have room to improve.
   • One of the pricier models per second in our quotes: $2.483 for 5 seconds at 720p.

3. Kling 3.0 Pro

   Ids t2v-kling-v3-pro, i2v-kling-v3-pro and f2v-kling-v3-pro for first and last frame. 3 to 15 seconds at 1080p or 4K. Quoted $0.706 for 5 seconds at 1080p.

   Best for: Multilingual dialogue scenes and multi-shot sequences you direct shot by shot

   Pros

   • Native speech in English, Chinese, Japanese, Korean and Spanish, including several characters speaking different languages (Kuaishou, February 2026).
   • Multi-shot direction such as shot-reverse-shot, with up to 6 shots in custom mode, and strong retention of text, signage and logos.
   • In FLORA: first-and-last-frame shape for transitions, and a low quote of $0.706 for 5 seconds at 1080p.

   Cons

   • Clips top out at 15 seconds.
   • Ranked mid-table, around 15th, on the Artificial Analysis leaderboards in October 2026, below the newest summer releases.

4. Veo 3.1, Veo 3.1 Fast and Veo 3.1 Lite

   Ids include i2v-veo3.1, i2v-veo3.1-fast and i2v-veo-3-1-lite-i2v, plus f2v and r2v versions. 4, 6 or 8 seconds. Quoted $3.84 (Veo 3.1) and $0.481 (Veo 3.1 Lite) for 8 seconds at 720p.

   Best for: Short, polished shots with native audio; Lite for low-cost drafts

   Pros

   • Native audio on every clip, first-and-last-frame control and up to 3 reference images (not on Lite).
   • 1080p and 4K output at 8 seconds on Veo 3.1 and Fast.
   • In FLORA: all three tiers side by side, from $0.481 (Lite) to $3.84 (Veo 3.1) for 8 seconds at 720p in our quotes.

   Cons

   • 8 seconds per clip; Google's docs say English prompts are best supported.
   • Still labeled preview in Google's API docs as of October 2026.

5. MiniMax H3 Max

   Ids include t2v-minimax-h3-max-gateway and i2v-minimax-h3-max-gateway, with first-last-frame, reference, video and audio versions. 1 to 15 seconds at 480P, 768P or 1080P. Quoted $0.48 for 5 seconds at 768P, with an estimated 30-second run.

   Best for: Fast, low-cost ad and ecommerce motion, including text and logos

   Pros

   • Top-ranked for image-to-video and in the top five for text-to-video on the Artificial Analysis leaderboards in October 2026.
   • MiniMax says it handles text and brand rendering, with native stereo audio, up to 15 seconds.
   • In FLORA: the cheapest default quote we saw, $0.48 for 5 seconds at 768P, with an estimated 30-second run.

   Cons

   • MiniMax says multimodal understanding and fine visual detail still need work; H3 Max has no 2K output.

6. Gemini Omni 1.1 Flash

   Ids include t2v-gemini-omni-1-1-flash-t2v and i2v-gengateway-omni-1-1-flash-gg, plus video-to-video and reference versions. 3 to 10 seconds, 360p to 4K, 16:9 or 9:16. Quoted $0.96 for 8 seconds at 720p.

   Best for: Short clips refined by editing, plus video-to-video changes

   Pros

   • Google's docs list conversational editing, extension, upscaling and interpolation, with output up to 4K.
   • In the top five for image-to-video on the Artificial Analysis leaderboards in October 2026.
   • In FLORA: text, image, video and reference shapes, quoted at $0.96 for 8 seconds at 720p.

   Cons

   • Up to 10 seconds per generation.

7. FLUX 3

   Ids t2v-flux-3-gateway, i2v-flux-3-gateway, v2v-flux-3-gateway and f2v-flux-3-gateway. 5 to 20 seconds at 720p or 1080p, with optional generated audio. Quoted $1.02 for 5 seconds at 720p.

   Best for: Stylized and typographic video with dialogue, up to 20 seconds

   Pros

   • Black Forest Labs lists text, image, video and keyframe inputs, up to 20 seconds with native audio and multilingual dialogue.
   • In the top six for text-to-video on the Artificial Analysis leaderboards in October 2026.
   • In FLORA: four input shapes, quoted at $1.02 for 5 seconds at 720p.

   Cons

   • Still in early access; Black Forest Labs says its evaluations are early and editing is coming later.

8. WAN 3.0 and WAN 3.0 Prime

   Ids t2v-wan-3-0-gateway (estimated $0.60, about 180 seconds) and t2v-wan-3-0-prime-gateway (estimated $0.84, about 45 seconds).

   Best for: Long text-to-video story clips at a moderate price

   Pros

   • Ranked first for text-to-video on the Artificial Analysis leaderboards in October 2026.
   • Alibaba lists up to 30 seconds, up to 1080p, native audio and partial re-edits without regenerating the whole clip.
   • In FLORA: two tiers, estimated at $0.60 (WAN 3.0) and $0.84 (Prime).

   Cons

   • Alibaba says audio texture and on-screen text accuracy are still maturing.
   • Unlike earlier WAN releases, WAN 3.0 is API-only, with no open weights announced.

What happened when we ran a video model in FLORA?

On October 16, 2026, we called flora_list_models with type video from Claude Code. It returned 299 video model entries. We then took one still, a 4:3 GPT Image 2.5 Flare image of a kraft coffee bag labeled NORTH FIELD ROAST on a cafe counter, and asked FLORA to quote an image-to-video clip from it on eight setups with flora_create_generations and quote set to true. Nothing was charged for the quotes.

We ran one of them: i2v-gengateway-seedance-2-5-i2v with duration 4, resolution 720p and aspect_ratio auto, and the motion prompt "Slow push-in toward the coffee bag on the counter. The printed label stays sharp and unchanged. In the soft background a barista moves behind the espresso machine, out of focus." It took 3 minutes 54 seconds and charged $1.996 against a $1.987 quote.

The clip is 4.09 seconds, 1112 x 834 at 24 fps, with an audio track. Auto kept the still's 4:3 framing. The label stays sharp and correctly spelled in every frame we checked, and a barista moves behind the counter, out of focus.

What went wrong: the push-in also drifted sideways, so the bag ends nearer the right edge than it started. Soft round highlights from the window also appear in the background late in the clip, which we did not ask for. For a straight move, write "straight push-in, no orbit, no lens flare".

Model run

i2v-gengateway-seedance-2-5-i2v, 4 s, 720p, auto

Tools called

flora_list_models, flora_create_generations (quote, then run), flora_list_generations

Cost

$1.996 charged; quotes were free

Time

3 min 54 s

Image-to-video or text-to-video: pros and cons

What do the video model id prefixes mean?

The prefix names the input shape. The same model family often has several, priced separately.

• t2v: text to video.
• i2v: one still image to video.
• f2v: first and last frame to video, for transitions.
• r2v: reference images to video, such as a character and a setting.
• v2v: an existing video to a new video, for restyling or edits.
• a2v, ai2v, av2v: audio-driven video, from audio alone, audio plus an image, or audio plus video.

Which video model should I start with?

Start from a still and use Seedance 2.5 image-to-video at 720p and a short duration while you work out the motion. Move up in resolution or length once the move is right. For quick motion tests, quote MiniMax H3 Max or Veo 3.1 Lite, which came in under $0.50 at their defaults.

How long does an AI video clip take?

Our Seedance 2.5 clips took 3 minutes 33 seconds for 5 seconds of video and 3 minutes 54 seconds for 4 seconds. FLORA's estimates range from about 30 seconds for MiniMax H3 Max to about 5 minutes for Seedance 2.5 and Veo 3.1. Let the agent poll in the background and keep working.

What else is in the video catalogue?

Earlier versions stay available, so you can match work made with an older model. The list includes Seedance 1.0 Pro, 1.5 Pro and 2.0, Kling 2.0 to 2.6 and Kling O1 and O3, Veo 2 and Veo 3, MiniMax Hailuo, WAN 2.2 to 2.7, LTX-2 Pro, Grok Imagine, HeyGen Video 1.0 and Tencent Hunyuan.

Frequently asked questions