Mochi 1: what the repository publishes

Mochi 1 publishes a 10 billion parameter model under Apache 2.0, with a separately stated 362M autoencoder. The repository asks for about 60GB of VRAM on a single GPU and recommends at least one H100, at 480p. As of 2026-09-12.

Inclusion rule. The same five fields for every family here. A field the repository does not state is recorded as not stated rather than estimated from the parameter count. Order. Fixed field order, identical on every family page.

1Permissive licence, demanding hardware, modest resolution

These three facts together describe a research release rather than a production drop-in. The licence imposes nothing, the hardware asks for a data-centre card, and the output resolution is the lowest stated in this register.

That combination is worth naming because the word open covers it and covers a 4GB model equally, and those two are not alternatives for the same team.

2Naming the recommended card is more useful than a bare figure

Sixty gigabytes narrows the field; recommending at least one H100 names it. A reader can price an H100 hour in a minute, and cannot price sixty gigabytes without first deciding which card supplies it.

Eight of the 15 families in this register name a card or a card generation. Those are the entries whose hardware cost can be estimated from published material alone.

3A stated present limit is not a permanent one

The repository describes 480p as the current capability, which is a form of words that says more is intended. This register records it as stated, with the date, and will change the row when the repository does.

Recording the qualifier matters here: a flat 480p row would read as a property of the model rather than as a snapshot of a release.

4The same field, across every family

Inclusion rule. Families whose vendor has published weights openly. A family whose repository does not state this particular field keeps its row, marked hollow. Order. Alphabetical by family name.

Inclusion rule. Families whose vendor has published weights openly. A family whose repository does not state this particular field keeps its row, marked hollow. Order. Alphabetical by family name.

Inclusion rule. Families whose vendor has published weights openly. A family whose repository does not state this particular field keeps its row, marked hollow. Order. Alphabetical by family name.

• Licence
  Apache 2.0stated in the repositoryGenmo, Mochi repository / recorded 2026-09-12
• Model size
  A 10 billion parameter diffusion model, with a 362M parameter video autoencoderstated in the repositoryGenmo, Mochi repository / recorded 2026-09-12
• Hardware
  Approximately 60GB VRAM on a single GPU, with at least one H100 recommendedthe vendor's own figureGenmo, Mochi repository / recorded 2026-09-12
• Resolution
  480p, described as the current capabilitystated as a present limitGenmo, Mochi repository / recorded 2026-09-12

5Running the model without the hardware

Instead of local hardware for “Mochi 1”, SceneMixer runs hosted models; its guides describe the result:

Seedance 2.0 (ByteDance's video model family) is one of the strongest engines for vertical short drama in 2026.

SceneMixer, Seedance 2.0 in a Real Short-Drama Pipeline (read 2026-10-13)

SceneMixer writes dialogue into the script analysis, assigns each line to a shot with a speaker and a short delivery note, and the video model performs it — picture and audio generated together, so the mouth forms that sentence in the language you chose.

SceneMixer, AI Dialogue Generator (read 2026-10-13)

6Sources

Read from github.com/genmoai on 2026-09-12. Hardware figures across every family are collected on requirements; the licence column is set out on licence and what counts as a published weight is on the method page. Previous family: CogVideoX. Next family: Allegro.