Mochi 1
video gen · reviewed by Stan K. · Source: data snapshot · data as of 2026-11-12
Apache 2.0 weights whose README says ~60 GB while the ComfyUI path runs under 20 GB on one card — plan around the path, not the README.
#9 of 14 in Video Gen
ToolScore breakdown — why 52/100
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
- ▸Apache 2.0 — no caps, no territories, no gating
- ▸The ComfyUI path brings it under 20 GB on a single card
- ▸Electricity economics once running: re-rolls cost watts
Where it costs you
- ▸The README number is ~60 GB — an order of magnitude above the practical path
- ▸Which hardware claim is true depends on whose instructions you follow
- ▸No length or resolution caps published
Hidden traps — the stuff the pricing page doesn't say
The official weights are ~60 GB; the ComfyUI path runs the same model under 20 GB on one card. The gap is not marketing spin — it is two different setups — but if you buy hardware against the README you will overspend, and if you plan against the ComfyUI number you are on a community-maintained path.
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
Data as of 2026-11-12. Prices and limits change monthly — report anything stale. How we verify: methodology.
Closest alternatives
The big Apache-licensed sibling with an official 80 GB target — not a consumer card; the 3060 path e
Consumer-reachable (14 GB min with offload) under Tencent's community license — NC in the EU, UK and
A clean MIT license from Meituan at 13.6B dense — and no verified hardware figure published yet.
Mochi 1 review
In-depth analysis · last updated 2026-11-12
Two numbers, one model
Mochi 1 ships under Apache 2.0 with two very different hardware stories. The README says ~60 GB of official weights — datacenter territory. The ComfyUI path brings the same model under 20 GB on one card. Both are true; they describe different setups. The trap is planning against the wrong one: budget GPUs for the README and you overspend, budget for the ComfyUI number and you are dependent on a community-maintained path.
What the license buys
Apache 2.0 is the cleanest terms on this page — no revenue cap, no territories, no gated weights. The license is not why this row scores mid-table; the operational confusion is. Once running, the economics are the local standard: ≈$0.03/hour of electricity for a 170 W card at the US average $0.18/kWh, and rejected takes re-roll for watts.
How the pillars land
Price/Value 3.5 — the official requirement prices out consumer rigs; the practical path exists but is not the supported one. Length & resolution caps 4.5 — no caps published; the hardware figure itself is contested between paths. Render speed 5 — neutral: no render-time figure in our snapshot. Training use & license 10 — Apache 2.0, nothing attached. API access 8 — open weights with a known single-card path via ComfyUI. Feature set 5 — neutral: no feature claims beyond the basics appear in our snapshot. Track record 3 — a README that far from the practical path is a project-maturity signal. Ease of use 3 — the gap bites the user first: which instructions you follow decides whether the model fits. Pricing transparency 5 — the sizes are published, confusingly; neutral. Support & docs 5 — neutral: no support-channel facts in our snapshot.
Verdict — 52/100
An Apache license attached to a confusing hardware story. If you already run ComfyUI and have 20 GB to spare, it works; if you are planning a rig around official numbers, this row will mislead you twice in opposite directions.