LBH-123-AI/Minimax_h3_latent_Upscaler

<p align="center"> <a href="/LBH-123-AI/Minimaxh3latentUpscaler/blob/main/README.md"><strong>English</strong></a> · <a href="/LBH-123-AI/Minimaxh3latentUpscaler/blob/main/README_zh.md">中文</a> </p>

<p align="center"> <a href="https://huggingface.co/LBH-123-AI/Minimaxh3latentUpscaler"><img src="https://img.shields.io/badge/dynamic/json?url=https%3A%2F%2Fhuggingface.co%2Fapi%2Fmodels%2FLBH-123-AI%2FMinimaxh3latentUpscaler&amp;query=%24.downloads&amp;label=downloads%20(last%2030d)&amp;color=blue&amp;logo=huggingface" alt="Downloads (last 30 days)"></a> <a href="https://huggingface.co/LBH-123-AI/Minimaxh3latentUpscaler"><img src="https://img.shields.io/badge/dynamic/json?url=https%3A%2F%2Fhuggingface.co%2Fapi%2Fmodels%2FLBH-123-AI%2FMinimaxh3latentUpscaler&amp;query=%24.likes&amp;label=likes&amp;color=red&amp;logo=huggingface" alt="Likes"></a> </p>

Minimax H3 Latent Upscaler

Neural latent-space upscaler for Minimax H3 video generation. Works directly on Minimax H3's 24-channel VAE latents to upscale spatial resolution (H×W) while preserving the time dimension.

Purpose

This model is designed to accelerate high-resolution H3 video generation:

1. 1.Generate video at low resolution — far fewer latent tokens, much faster.
2. 2.Upscale the latent in-place with this learned upscaler.
3. 3.Re-sample / refine at the target resolution to recover detail.

By skipping the expensive decode → pixel upscale → encode round-trip through Minimax H3's heavy ~5B-parameter VAE, this pipeline saves a significant amount of generation time. It also avoids the ghosting / double-image artifacts that naive latent interpolation (bilinear/bicubic) introduces.

📸 Examples

Video upscale comparison — click to play:

<video src="https://huggingface.co/LBH-123-AI/Minimaxh3latentUpscaler/resolve/main/examples/Minimaxh3latentUpscaler_001.mp4" controls width="640"></video>

(If the player doesn't render, [download the video here](https://huggingface.co/LBH-123-AI/Minimax_h3_latent_Upscaler/resolve/main/examples/Minimax_h3_latent_Upscaler_001.mp4).)

Image upscale comparison:

Files in this repository

The current release is v1. Its three checkpoints live together in `minimax_h3_latent_upscaler_3d_conv_v1/`:

All three checkpoints share the same 3D-convolution architecture. Pick the precision that matches your GPU and workflow.

config.json at the repository root is the family index: it records what every release shares (the 24-channel H3 latent space and its normalization, the supported scale range, the ComfyUI node mapping) and points to each released variant. The architecture spec for this release sits next to the weights, in `minimax_h3_latent_upscaler_3d_conv_v1/config.json`: 24 in/out channels, 512 base channels, 12+12 blocks, temporal conv every 2 blocks with kernel size 5, 345,280,216 parameters.

The ComfyUI node infers the architecture from the state dict on load and does not read either file — they are machine-readable specifications for downstream tooling.

Versioning

Each release gets its own folder. New checkpoints will go to minimax_h3_latent_upscaler_3d_conv_v2/, ..._v3/, and so on, so older releases stay downloadable and unchanged.

Filenames always carry the full release tag. The folder groups a release together for browsing, but the filename is what actually identifies a checkpoint once it is on your disk — ComfyUI loads every model from a single flat directory (models/latent_upscale_models/).

Usage

Use with the companion ComfyUI custom node:

[LBH-123-AI/Comfyui_Minimax_h3_latent_Upscaler](https://github.com/LBH-123-AI/Comfyui_Minimax_h3_latent_Upscaler)

1. 1.Install the custom node into ComfyUI/custom_nodes/.
2. 2.Place the downloaded checkpoint in:

   ComfyUI/models/latent_upscale_models/

1. 1.Add either "Minimax H3 Latent Upscaler (2D)" or "Minimax H3 Latent Upscaler (3D)" from the video/MinimaxH3 menu, connect a LATENT, pick the model, and set scale.

Supported upscale factors: 1.0× – 4.0× (continuous, 0.1 step; default 2.0×).

Training Data

The model was trained on ~80,000 paired samples (low-resolution latent + high-resolution target), balanced across modalities and scale factors to maximize generalization.

• —~70,000 video pairs
• —~8,000 2K image pairs

Scale distribution (approximate):

Architecture

3D-convolution backbone with temporal convolution and trilinear interpolation. The architecture draws on and references the LTX 2.3 Spatial Upscaler (ltx-2.3-spatial-upscaler-x2-1.1.safetensors) and the neural-latent-upscaling approach pioneered by Ttl / ComfyUi_NNLatentUpscale.

License

Apache-2.0