H3-to-LTX-Latent-Adapter: 4GB+ VRAM, File List
At a Glance
| Item | Value |
|---|---|
| Repository | Efficient-Large-Model/H3-to-LTX-Latent-Adapter |
| Publisher guide | Efficient Large Model (NVIDIA and MIT): models and licenses |
| Published | 2026-09-10 |
| Formats | safetensors |
| Source type | Primary source (the publisher itself) |
Values determined by this site’s code when the information was collected. Dates are JST.
Overview
An inference adapter named “H3-to-LTX Latent-Adapter" has been released for converting MiniMax-H3 video latents into LTX-2.5 Conv VideoVAE latents in BF16. This model serves to transform normalized MiniMax-H3 video latents into normalized LTX-2.5 latents.
This model does not possess the ability to generate videos on its own, and it does not include the MiniMax-H3 or LTX model weights themselves. It is primarily intended for use in Sol-H3-Spark within the Sana repository as a component to connect different models.
Specifications
The technical specifications of this model are as follows:
- Architecture: 194,759,504-parameter Conv3D residual adapter (width 752, 22 blocks)
- Conversion Process: Applies publicly available temporal/spatial packing transforms
- Input Specification: Normalized H3 latent
[B, 24, T_h3, H/16, W/16] - Output Specification: Normalized LTX latent
[B, 128, T_ltx, H/32, W/32] - Data Type: BF16
- Distribution Format: safetensors
- Configuration File:
config.json(defines architecture, geometry, and normalization contracts) - Weights SHA-256:
170199a390c40ac97f5895bc9c8cc29817e74fb9193c858a85d8c0f1f30724ac
During inference, it is necessary to pass the original pixel-frame count and pixel geometry to the adapter’s convert method. Based on this, the Sana inference library performs the appropriate packing and normalization. Note that modifying this conversion process (transform) is not compatible with this checkpoint.
Regarding the license, no new weight license is assigned by this adapter. The MiniMax-H3 and LTX components used alongside the adapter are reported to continue following their respective terms of use.
Strengths and Use Cases
This model is an adapter specialized in converting MiniMax-H3 video latents into LTX-2.5 Conv VideoVAE latents. It does not have the ability to generate videos by itself, nor does it contain the MiniMax-H3 or LTX model weights. Therefore, it functions as a “bridge" to connect video generation models with different architectures.
As a specific intended use case, its utilization in Sol-H3-Spark within the Sana repository is mentioned. By performing conversions with appropriate packing and normalization from the MiniMax-H3 latent space to the latent space handled by LTX-2.5, it enables video generation workflows combining these two models. The tags specify minimax-h3, ltx-2.5, latent-adapter, and video, indicating that it is a tool aimed at latent conversion in video generation.
Hardware Requirements
Estimated requirements (calculated by Local Model Watch) — 195M parameters
| Your VRAM | Quantization | File size | Est. memory needed |
|---|---|---|---|
| 4GB (laptop iGPU / phone class) | BF16 | 0.4GB | 0.4GB |
Memory estimates add a 20% runtime overhead (KV cache, etc.) to the actual size of the distributed files. Actual usage varies with context length, batch size and inference engine. These figures are computed by this site from file sizes, not published by the model’s authors. Compare with other models in our VRAM quick reference. What the quantization names mean: glossary.
Our Own Measurements
Measurements We Did Not Take
This model does not state a license, so we cannot confirm permission to use it. Because this site carries advertising, we did not run the model (no CPU run, answers, quantization comparison, conversion or generation).
Distributed Files
Weight files published in Efficient-Large-Model/H3-to-LTX-Latent-Adapter, listed by this site from the Hugging Face API. Sizes are the actual file sizes.
| File | Size |
|---|---|
model.safetensors |
390MB |
How to Get It
This model is available from the Hugging Face repository Efficient-Large-Model/H3-to-LTX-Latent-Adapter.
The distribution format is safetensors, and it supports inference in BF16 precision. The contracts concerning the model’s architecture, geometry, and normalization are described in config.json.
For technical usage, it is necessary to pass the original pixel-frame count and pixel geometry to the adapter’s convert method. The Sana inference library executes appropriate packing and normalization based on this. Note that altering this conversion process (transform) is not compatible with this checkpoint.
The MiniMax-H3 and LTX components used alongside this adapter are reported to continue following their respective terms of use.
Related Articles
- LongLive-Plug-Wan2.1-T2V-14B-few-step: Our Generated Video, File List
- LongLive-Plug-Wan2.1-T2V-14B-cfg: Our Generated Video, File List
- LongLive-Plug-Wan2.2-TI2V-5B-cfg: Our Generated Video, File List
- LongLive-Plug-Wan2.2-TI2V-5B-few-step: Our Generated Video, File List
What to Read Next
- Find models by VRAM (This model runs from the 4GB tier) → Other models that run on a 8GB GPU
- Formats this model is available in → Safetensors format guide and models
- Learn about the publisher → Efficient Large Model (NVIDIA and MIT): models, licenses and articles
Sources
Update History
- 2026-09-20: Rewrote the article from re-collected sources and restored it from draft to published.
- 2026-09-25: Changed the title to show what the article covers (VRAM requirements, file list, etc.).