NodeValid

Published video-model weights, licences and hardware notes

MAGI-1: what the repository publishes

MAGI-1 states hardware as machine shapes rather than memory: H100 or H800 times eight for the 24B model, one RTX 4090 for the 4.5B. It adds that the 4.5B needs at least 24GB, and that a distilled fp8 build runs on 12GB. The licence is Apache 2.0. As of 2026-09-22.

Hardware written as a count of cardsThe model table answers in machines rather than gigabytes: eight H100 or H800 cards for the 24B release, one RTX 4090 for the 4.5B, and a distilled fp8 build that is said to work on 12GB. A count of a named card converts into a rented price without any further decision.Which release is being runMAGI-1-24BH100 or H800, eight ofthemA count of data-centrecards rather than a memoryfloor.MAGI-1-4.5BOne RTX 4090At least 24GB is describedas sufficient for thissize.4.5B distilled, fp8At least 12GBA figure that belongs tothis build rather than tothe model.The unit is a machine, not a number
Fig. 1 Three answers to one question, and each names the machine rather than a number the reader has to turn into a machine.
What the repositories actually stateFilled where the repository states something, hollow where it does not. A blank is not an inference.What the repositories actually stateWhat is statedLicenceLicence — What is stated: Apache License 2.0SizeSize — What is stated: 24B and 4.5B modelsHardwareHardware — What is stated: H100 or H800 x 8 for the 24BResolutionResolution — What is stated: 720 x 720 as the 4.5B defaultLengthLength — What is stated: No ceiling recorded
Fig. 2 Filled where the repository states something, hollow where it does not. A blank is not an inference.
The five fields this register keeps, for MAGI-1. Read from the vendor's own repository on 2026-09-22.
FieldWhat is statedDetail
LicenceApache License 2.0Stated in the repository
Size24B and 4.5B modelsBoth published, each with its own hardware note
HardwareH100 or H800 x 8 for the 24BOne RTX 4090 for the 4.5B; 12GB for a distilled fp8 build
Resolution720 x 720 as the 4.5B defaultSet for speed; arbitrary resolutions described as supported
LengthNo ceiling recordedFrames are an argument with no stated maximum

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.

1Eight cards of one kind, or one card of another

The hardware column of the model table reads H100/H800 x 8 and RTX 4090 x 1. That is a count of a named card, which is the one hardware unit in this register that converts into a rented price without an intervening decision by the reader.

It also describes the shape of the problem rather than a threshold. A 24B model wanting eight data-centre cards is not a model that one larger card unlocks, and no memory figure would have carried that as directly as a multiplier does.

2Two floors for the smaller model, and the lower one belongs to a build

Any machine with at least 24GB of GPU memory is described as sufficient for the 4.5B models, and a distilled fp8 configuration with a window size of one is said to work on GPUs with at least 12GB. The second figure is a property of that specific build, not of the model.

Halving a memory floor by distilling and quantising is a routine move, and this repository is unusual in publishing the result of it as a number. Elsewhere in this register the same move shows up only as the existence of a variant, with the hardware left for a reader to discover by trying.

3A resolution stated as a default chosen for speed

The default resolution for the 4.5B model is 720 by 720, and the repository says it was set that way to accelerate inference. Arbitrary resolutions are described as supported, so the figure is a starting point rather than a capability limit.

No clip length is stated anywhere. Frames are passed as an argument and no maximum is attached to them, so the length cell is empty on the entry whose hardware cell is the fullest in this register.

4The same field, across every family

Licence, family by familyOne field at a time, every family beside it, MAGI-1 marked. Quoted from each repository and never derived from a parameter count. The exact phrasing sits in the table under this figure. Recorded 2026-09-22.Licence, family by familyLicenceAllegroAllegro — Licence: apache-2.0CogVideoXCogVideoX — Licence: Split by model sizeCosmos Predict 2.5Cosmos Predict 2.5 — Licence: NVIDIA Open Model LicenseFramePackFramePack — Licence: Apache 2.0HunyuanVideoHunyuanVideo — Licence: tencent-hunyuan-communityLongCat-VideoLongCat-Video — Licence: MIT LicenseLTX-VideoLTX-Video — Licence: Apache 2.0MAGI-1MAGI-1 — Licence: Apache License 2.0Mochi 1Mochi 1 — Licence: Apache 2.0Open-SoraOpen-Sora — Licence: Apache 2.0Pyramid FlowPyramid Flow — Licence: stabilityai-ai-communitySkyReels-V2SkyReels-V2 — Licence: skywork-licenseStable Video DiffusionStable Video Diffusion — Licence: stable-video-diffusion-communityStep-Video-T2VStep-Video-T2V — Licence: MITWanWan — Licence: Apache 2.0
Fig. 3 One field at a time, every family beside it, MAGI-1 marked. Quoted from each repository and never derived from a parameter count. The exact phrasing sits in the table under this figure. Recorded 2026-09-22.
Every family on one field, with MAGI-1 marked. Recorded 2026-09-22.
FamilyLicence
Allegroapache-2.0
CogVideoXSplit by model size
Cosmos Predict 2.5NVIDIA Open Model License
FramePackApache 2.0
HunyuanVideotencent-hunyuan-community
LongCat-VideoMIT License
LTX-VideoApache 2.0
MAGI-1Apache License 2.0
Mochi 1Apache 2.0
Open-SoraApache 2.0
Pyramid Flowstabilityai-ai-community
SkyReels-V2skywork-license
Stable Video Diffusionstable-video-diffusion-community
Step-Video-T2VMIT
WanApache 2.0

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.

Size, family by familyOne field at a time, every family beside it, MAGI-1 marked. Quoted from each repository and never derived from a parameter count. The exact phrasing sits in the table under this figure. Recorded 2026-09-22.Size, family by familySizeAllegroAllegro — Size: VAE 175M and DiT 2.8BCogVideoXCogVideoX — Size: 2B and 5B, carried in the model namesCosmos Predict 2.5Cosmos Predict 2.5 — Size: 2,059,174,912 parametersFramePackFramePack — Size: 13B modelsHunyuanVideoHunyuanVideo — Size: Over 13 billion parametersLongCat-VideoLongCat-Video — Size: 13.6B parametersLTX-VideoLTX-Video — Size: A 13B modelMAGI-1MAGI-1 — Size: 24B and 4.5B modelsMochi 1Mochi 1 — Size: 10 billion parametersOpen-SoraOpen-Sora — Size: An 11B modelPyramid FlowPyramid Flow — Size: No parameter count recordedSkyReels-V2SkyReels-V2 — Size: 1.3B, 5B and 14B variantsStable Video DiffusionStable Video Diffusion — Size: 2B paramsStep-Video-T2VStep-Video-T2V — Size: 30 billion parametersWanWan — Size: A 5B text-to-video and image-to-video model
Fig. 4 One field at a time, every family beside it, MAGI-1 marked. Quoted from each repository and never derived from a parameter count. The exact phrasing sits in the table under this figure. Recorded 2026-09-22.
Every family on one field, with MAGI-1 marked. Recorded 2026-09-22.
FamilySize
AllegroVAE 175M and DiT 2.8B
CogVideoX2B and 5B, carried in the model names
Cosmos Predict 2.52,059,174,912 parameters
FramePack13B models
HunyuanVideoOver 13 billion parameters
LongCat-Video13.6B parameters
LTX-VideoA 13B model
MAGI-124B and 4.5B models
Mochi 110 billion parameters
Open-SoraAn 11B model
Pyramid FlowNo parameter count recorded
SkyReels-V21.3B, 5B and 14B variants
Stable Video Diffusion2B params
Step-Video-T2V30 billion parameters
WanA 5B text-to-video and image-to-video model

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.

Hardware, family by familyOne field at a time, every family beside it, MAGI-1 marked. Quoted from each repository and never derived from a parameter count. The exact phrasing sits in the table under this figure. Recorded 2026-09-22.Hardware, family by familyHardwareAllegroAllegro — Hardware: 9.3G BF16 with cpu_offloadCogVideoXCogVideoX — Hardware: From 4GB in FP16 on the 2BCosmos Predict 2.5Cosmos Predict 2.5 — Hardware: 32.54 GB of GPU VRAMFramePackFramePack — Hardware: 6GB minimum for 1800 frames at 30fpsHunyuanVideoHunyuanVideo — Hardware: 60GB peak at 720px1280px, 129 framesLongCat-VideoLongCat-Video — Hardware: No memory figure recordedLTX-VideoLTX-Video — Hardware: No minimum statedMAGI-1MAGI-1 — Hardware: H100 or H800 x 8 for the 24BMochi 1Mochi 1 — Hardware: About 60GB VRAM on a single GPUOpen-SoraOpen-Sora — Hardware: 60.3GB peak at 768px on one GPUPyramid FlowPyramid Flow — Hardware: No memory figure recordedSkyReels-V2SkyReels-V2 — Hardware: About 14.7GB peak on the 1.3B at 540PStable Video DiffusionStable Video Diffusion — Hardware: About 180s on an A100 80GB cardStep-Video-T2VStep-Video-T2V — Hardware: 78.55 GB peak at 768px768px204fWanWan — Hardware: A consumer 4090 on the 5B release
Fig. 5 One field at a time, every family beside it, MAGI-1 marked. Quoted from each repository and never derived from a parameter count. The exact phrasing sits in the table under this figure. Recorded 2026-09-22.
Every family on one field, with MAGI-1 marked. Recorded 2026-09-22.
FamilyHardware
Allegro9.3G BF16 with cpu_offload
CogVideoXFrom 4GB in FP16 on the 2B
Cosmos Predict 2.532.54 GB of GPU VRAM
FramePack6GB minimum for 1800 frames at 30fps
HunyuanVideo60GB peak at 720px1280px, 129 frames
LongCat-VideoNo memory figure recorded
LTX-VideoNo minimum stated
MAGI-1H100 or H800 x 8 for the 24B
Mochi 1About 60GB VRAM on a single GPU
Open-Sora60.3GB peak at 768px on one GPU
Pyramid FlowNo memory figure recorded
SkyReels-V2About 14.7GB peak on the 1.3B at 540P
Stable Video DiffusionAbout 180s on an A100 80GB card
Step-Video-T2V78.55 GB peak at 768px768px204f
WanA consumer 4090 on the 5B release

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 License 2.0stated in the repositorySandAI-org, MAGI-1 repository / recorded 2026-09-22
  • Model size
    MAGI-1-24B and MAGI-1-4.5Btwo published families of weightsSandAI-org, MAGI-1 repository / recorded 2026-09-22
  • Hardware
    H100/H800 x 8 for the 24B model, RTX 4090 x 1 for the 4.5Bstated as a count of a named cardSandAI-org, MAGI-1 repository / recorded 2026-09-22
  • Hardware
    For 4.5B models, any machine with at least 24GB of GPU memory is sufficient, and a distilled fp8 configuration works on GPUs with at least 12GBa floor for the model and a lower one for a buildSandAI-org, MAGI-1 repository / recorded 2026-09-22
  • Resolution
    The default resolution for the 4.5B model is set to 720x720a default chosen to accelerate inferenceSandAI-org, MAGI-1 repository / recorded 2026-09-22

5Sources

Read from github.com/SandAI-org on 2026-09-22. 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 how read. Previous family: SkyReels-V2. Next family: Cosmos Predict 2.5.