Wan 2.2 TI2V-5B
video gen · reviewed by Stan K. · Source: data snapshot · data as of 2026-11-12
The consumer pick of local video: Apache 2.0, an official 24 GB target and under 9 minutes per 720p 24fps clip — the only local row with verified official hardware numbers.
#2 of 14 in Video Gen
ToolScore breakdown — why 88/100
10 criteria × 10 points. Low pillars are visible on purpose — that's the point of a score you can audit. Methodology →
What it does well
- ▸Official 24 GB target — a consumer card, not a datacenter reservation
- ▸Under 9 minutes per 720p 24fps clip on a consumer GPU — the only local row with a published render time
- ▸Apache 2.0 license: no revenue cap, no territories, no gating
- ▸Re-rolls cost watts, not credits — the keeper-rate multiplier that punishes paid APIs does not apply
- ▸Same family as Wan 3.0, the current AA Video Elo king (1175)
Where it costs you
- ▸Below 24 GB you leave the official path: 12 GB works only via community GGUF Q4/Q5 quants
- ▸No verified 3060 report exists — the 12 GB path is community territory, not a claim we can make
- ▸5B is the consumer scale of the family — the 14B sibling wants 80 GB
Hidden traps — the stuff the pricing page doesn't say
On 12 GB the model exists only as community GGUF Q4/Q5 quantizations. No verified RTX 3060 report exists for video workloads — treat 'it runs on a 3060' as a hope, not a spec.
The official target is 24 GB of VRAM. Anything below means offloading and quantization on your own — performance and quality claims there are community, not vendor, territory.
Know the limits before you pay: read the traps above, then decide with eyes open.
Affiliate link where available — never affects the score.
- internal ToolScore Video Gen hub — snapshot 2026-11-12 (numbers from LICENSE files and model cards) verified
- benchmark AA Video Leaderboard — Wan 3.0 at Elo 1175 (quality reference, 2026-11-12) verified
Data as of 2026-11-12. Prices and limits change monthly — report anything stale. How we verify: methodology.
Closest alternatives
The honest API: pay-as-you-go per second with no subscription at all — HD $0.75/5s, Draft HD $0.30/5
Veo's successor — the cheapest verified 720p-with-audio clip on the page (≈$0.35/5s) with published
The only local model with native audio+video in one pass — voice and picture from a single 19B model
Wan 2.2 TI2V-5B review
In-depth analysis · last updated 2026-11-12
The local default
Wan 2.2 TI2V-5B is the consumer pick of local video generation, and it earns that on two published numbers no other local row can match: an official 24 GB VRAM target — a consumer card, not a datacenter reservation — and a render time under 9 minutes per 720p 24fps clip on a consumer GPU. The license is Apache 2.0: no revenue cap, no territory exclusions, no gated weights. In a table where licenses are the main scoring hazard, this row has none.
The economics of watts
There is no per-unit price here — the cost model is electricity. A 170 W card draws ≈$0.03/hour at the US average rate ($0.18/kWh), and the number that makes local video economics work is the re-roll: a rejected clip costs another nine minutes of watts, not another credit. Paid APIs bill every generated clip; here the keeper-rate multiplier that punishes per-click pricing simply does not apply. The family pedigree helps too: the AA Video Elo king is Wan 3.0 at 1175 (snapshot 2026-11-12) — the same family whose Wan 2.2 release runs locally under Apache 2.0.
The 12 GB frontier
Below 24 GB you are off the official path. On 12 GB the model exists only as community GGUF Q4/Q5 quantizations — a community quant, with no verified 3060 report. Our hub refuses to claim one, and so does this page: the 3060 path is plausible, unverified, and entirely community-owned.
How the pillars land
Price/Value 10 — electricity-only economics with free re-rolls; the strongest value-per-watt in the local table. Length & resolution caps 9 — 720p 24fps clips at the official target, the target itself published. Render speed 9 — under 9 min per 720p clip is a published render-time fact, unique among local rows. Training use & license 10 — Apache 2.0, nothing attached. API access 9.5 — no hosted API; access is the weights plus community quants, the maximum form of programmatic access. Feature set 8.5 — text+image to video at the consumer scale of the Elo-king family. Track record 8.5 — the family ships the current quality king. Ease of use 9 — an official consumer-GPU target you can plan hardware around. Pricing transparency 9 — the hardware target and render time are published. Support & docs 5 — neutral: no support-channel facts appear in our snapshot.
Verdict — 88/100, BEST VALUE (local)
If you own a 24 GB-class card, this is the local answer: verified official numbers, a clean license, and rejects that cost cents of electricity instead of dollars of credits. Below 24 GB, downgrade your expectations with your VRAM — the quant path is real but officially unsupported.