# Vi · Video — element 20 of 58

> Motion pictures from prose. Turns scripts into watchable video.

- **Group:** 5 · Creative & Design
- **Necessity:** Optional
- **Price band:** $$ · $30–150/mo
- **Maturity:** Emerging
- **Edition:** v2026.Q3 · verified 2026-09-13

## Leading tools (v2026.Q3)

- **Veo (3.1 / Flow)** — quality plus everywhere
- **Seedance 2.5** — arena-topping storyteller
- **Runway (Gen-4.5)** — the editor's toolkit
- **Kling 3.0** — cinema per dollar
- **Hailuo (MiniMax H3)** — 2k crowd favorite

## Our take

Improving faster than any other creative element. Budget experiments, not commitments.

## Combines with

Im, Av, Vo


## The top 5 — deep dossier (verified 2026-09-13)

Veo, if you can only hold one — Google's stack is the only one that couples near-top quality (native audio, 4K, ~148s extendable shots) with real distribution: Gemini, Flow, a clean API, and resale inside Runway and every aggregator, plus the #1 crowd-arena model (Gemini Omni Flash, 1,244 Elo) in the same family. Seedance is the pick when story beats clips — 30-second single generations with multi-reference control, and the top image-to-video Elo — if you can live with its IP-driven availability wobble. Runway wins when you edit as much as you generate: Gen-4.5's minute-long character-consistent multi-shot plus Aleph and a pro suite no model-only vendor matches. Kling 3 is the value cinema play at ~$0.11/s with native audio. Hailuo (MiniMax H3) is the crowd's #2-rated model with native 2K — the budget wildcard. Sora, the tool that defined the category's hype cycle, is dead: app discontinued April 26, 2026, API off September 24.

1. **Veo (3.1 / Flow)** (Google DeepMind) — Google AI Pro $19.99/mo (1,000 Flow credits) · Ultra $99.99–199.99/mo (10k–25k credits) · API ~$0.10–0.40/s 720p–1080p, up to $0.60/s 4K+audio (via fal, Aug 2026). Best for: The default: best blend of quality, native audio, control features, and sheer availability — in Gemini, Flow, the API, and inside rivals' own products. Why: Veo 3.1 generates 1080p/4K with native dialogue, SFX and ambient audio, camera controls, character consistency, and scene extension to ~148 seconds — and it ships everywhere, including inside Runway's own paid tiers. Google also holds the #1 spot on the Artificial Analysis text-to-video arena outright with Gemini Omni Flash (1,244 Elo, May 2026), so the quality crown stays in-house even as Veo 3.1 itself ages. Watch: Veo 3.1 has slid to mid-table on the crowd arena (#11, 1,093 Elo) as Chinese models passed it — Google's answer (Omni Flash) lives in Gemini, not the Veo API, muddying the lineup. Base generations are still 8s clips, and 4K-with-audio at $0.60/s adds up fast. [https://deepmind.google/models/veo/](https://deepmind.google/models/veo/)
2. **Seedance 2.5** (ByteDance Seed) — Dreamina free daily credits · API usage-based via BytePlus/fal/Replicate (2.0 Pro/Fast/Mini tiers) · unlimited via partners (e.g. Higgsfield, from Aug 7, 2026). Best for: Actual storytelling — 30-second single generations, native multi-shot, and multi-reference control (up to 9 images, 3 clips, 3 audio files) instead of stitched 8-second fragments. Why: Seedance 2.0 holds #1 on the image-to-video arena (1,197 Elo) and #3 on text-to-video (1,224, Mar 2026), with 1.3M+ Replicate runs; 2.5 extends single generations to 30 seconds (twice extendable) with green-screen editing, camera blocking, and reference-video interpretation. No other model turns a script into a coherent multi-shot sequence with this little glue work. Watch: The IP overhang is real: Disney cease-and-desists over celebrity/character output forced ByteDance to pause the 2.0 global launch on Mar 15, 2026, and first-party pricing remains opaque — access flows through Dreamina, aggregators, and partner resellers. TikTok-adjacent geopolitics is a standing platform risk. [https://seed.bytedance.com/en/seedance](https://seed.bytedance.com/en/seedance)
3. **Runway (Gen-4.5)** (Runway AI) — Free 125 one-time credits · Standard $12/mo (annual; $15 monthly, 625 credits) · Pro $28/35 (2,250) · Max $76/95 (9,500) · Enterprise custom. Best for: Founders producing finished pieces, not clips — generation plus editing (Aleph), performance transfer (Act-Two), upscaling, and now other vendors' models in one workflow. Why: Gen-4.5 (Dec 2025 update) does native audio and minute-long multi-shot generation with character consistency, and the surrounding toolchain — Aleph edit-anything, Topaz 4K upscale, custom voices — is what turns raw generations into watchable video. The company raised $315M at $5.3B (Feb 10, 2026) and hedged model risk by hosting Veo 3.1 and others, adding a model router in July 2026. Watch: Runway doesn't enter the crowd arena, and its 'world's best video model' claim dates to the Nov 2025 launch — on raw generation quality the Chinese labs have likely passed it. Credit economics bite at scale, and the strategic center of gravity is drifting to world models (GWM-1) over creator video. [https://runwayml.com](https://runwayml.com)
4. **Kling 3.0** (Kuaishou) — API ~$0.112/s · $0.168/s with audio · $0.196/s with voice control (via fal, Aug 2026); consumer tiers on kling.ai (page JS-rendered — see notes). Best for: The best price-to-cinema ratio — 15-second 1080p clips with native audio and voice control at roughly a dollar per shot. Why: Kling 3.0 (Feb 2026) sits top-10 on the text-to-video arena in three variants (Pro 1080p at 1,111 Elo) and generates up to 15s with native audio, lip-sync, and a voice-control tier — at $0.112–0.196/s, roughly half Veo's standard rate. For volume experimentation it is the strongest cost-quality point among the closed leaders. Watch: First-party global pricing is hard to even read (JS-walled site), documentation trails the US vendors, and Kling is one product inside a short-video giant whose priorities are its own feed. Elo sits a tier below Seedance/Omni Flash/H3. [https://kling.ai](https://kling.ai)
5. **Hailuo (MiniMax H3)** (MiniMax) — Consumer subs via hailuoai.video (page JS-rendered — see notes) · API usage-based via MiniMax platform / fal / Replicate, standard + fast tiers. Best for: Budget wildcard with crowd-validated quality — the #2-rated model overall, native 2K output, and cheap fast variants for iteration. Why: H3 (Jul 2026) rates 1,238 Elo — #2 on text-to-video, within 6 points of Google's leader — and #3 on image-to-video, generating native 2K from text or images with omni-style multimodal context. MiniMax is now a public company (HKEX IPO Jan 9, 2026), taking the going-dark risk off the table. Watch: 2025 revenue was just $79M against frontier-scale compute burn — the subsidized pricing may not hold. The ecosystem is thin (no editing suite, patchy docs in English), and neither the consumer nor first-party API price list could be verified from vendor pages. [https://hailuoai.video](https://hailuoai.video)

### How to choose
- If You can only hold one subscription and need reliable quality with audio, everywhere → Veo via Google AI Pro ($19.99/mo, 1,000 Flow credits) — accept 8s base clips and extend; move to the API when volume justifies $0.10–0.40/s.
- If The job is a story — ads, trailers, episodic shorts — not isolated clips → Seedance 2.5: 30s single generations with multi-reference control beat stitching 8s fragments, if you accept the IP-friction availability risk and reseller-mediated pricing.
- If You edit as much as you generate, or deliver client-ready pieces → Runway — Gen-4.5's minute-long consistent multi-shot plus Aleph/Act-Two/upscaling is the only real post-production stack, and its model router hedges the generation layer.
- If Cost per watchable second decides, and you iterate in volume → Kling 3.0 (~$0.11/s, audio $0.168/s) or Hailuo fast tiers — the arena says you give up little quality for roughly half Veo's rate.
- If You need open weights — on-prem, fine-tuning, or no per-second meter → LTX-2.3 (22B open weights, 20s joint audio-video, free under $10M revenue) or the Wan 2.2 Apache lineage; accept a quality tier below the closed leaders.

### The field (18 more)

Gemini Omni Flash, Wan, LTX-2.3, Luma Dream Machine (Ray3.14), Pika 2.5 (fading), Vidu Q3, PixVerse v5.6, Grok Imagine Video 1.5, SkyReels V4, HappyHorse 1.1, MAGI-2 Preview, Moonvalley Marey, Higgsfield, Hunyuan Video (fading), Adobe Firefly Video, Mochi 1 (fading), Stable Video Diffusion (dead), Sora / Sora 2 (dead)

Full dossier data: https://elems.ai/e/vi.json

---
Source: [elems.ai](https://elems.ai/e/vi.html) — the periodic table of the AI-led startup. Data: https://elems.ai/elements.json (CC BY 4.0, cite elems.ai).