---
title: BennyDaBall-LTX-2.5-22b-distilled-nvfp4-comfy
canonical_url: "https://www.modelscope.cn/models/hf/BennyDaBall-LTX-2.5-22b-distilled-nvfp4-comfy"
md_url: "https://www.modelscope.cn/models/hf/BennyDaBall-LTX-2.5-22b-distilled-nvfp4-comfy.md"
repository: hf/BennyDaBall-LTX-2.5-22b-distilled-nvfp4-comfy
last_updated: 2026-10-16
license: other
pipeline_tag: text-to-video-synthesis
tasks:
  - text-to-video-synthesis
base_model:
  - Lightricks/LTX-2.5
base_model_relation: quantized
parameters: 13.9B
tensor_type:
  - BF16
  - F32
  - F8_E4M3
  - U8
library_name:
  - safetensors
  - pytorch
frameworks:
  - pytorch
downloads: 78
stars: 0
tags:
  - ltx
  - ltx-2.5
  - nvfp4
  - comfyui
  - diffusion-single-file
  - quantized
  - text-to-video
  - image-to-video
  - video-generation
  - audio-video
---

# BennyDaBall-LTX-2.5-22b-distilled-nvfp4-comfy

> BennyDaBall-LTX-2.5-22b-distilled-nvfp4-comfy - hf 在 ModelScope 开源的模型。LTX-2.5 22B Distilled — NVFP4, ComfyUI-ready The full NVFP4 build of the LTX-2.5 22B distilled transformer, stamped so ComfyUI actually loads it. Smallest working 22B on the board —…

hf/BennyDaBall-LTX-2.5-22b-distilled-nvfp4-comfy 是 ModelScope 魔搭社区上的 13.9B 参数text-to-video-synthesis模型，采用 other 许可，基于 Lightricks/LTX-2.5 构建。

- **Repository**: hf/BennyDaBall-LTX-2.5-22b-distilled-nvfp4-comfy
- **License**: other
- **Tasks**: text-to-video-synthesis
- **Parameters**: 13.9B
- **Base model**: Lightricks/LTX-2.5
- **Tags**: ltx, ltx-2.5, nvfp4, comfyui, diffusion-single-file, quantized, text-to-video, image-to-video, video-generation, audio-video
- **Downloads**: 78
- **Stars**: 0
- **Last updated**: 2026-10-16

Source: https://www.modelscope.cn/models/hf/BennyDaBall-LTX-2.5-22b-distilled-nvfp4-comfy

---

# LTX-2.5 22B Distilled — NVFP4, ComfyUI-ready

![LTX-2.5 22B Distilled NVFP4, ComfyUI-ready, by BennyDaBall](banner.png)

The full NVFP4 build of the LTX-2.5 22B distilled transformer, stamped so ComfyUI actually loads it. Smallest working 22B on the board — **18.7GB, ~2.8GB lighter than the official comfy-int8** — and the same-seed output holds up frame for frame.

<p align="center"><img src="https://huggingface.co/BennyDaBall/LTX-2.5-22b-distilled-nvfp4-comfy/resolve/main/assets/nvfp4_sign_hero.webp" alt="NVFP4 in ComfyUI" width="800"></p>
<p align="center"><i>Image-to-video with this checkpoint, 50fps — the sign text stays legible the whole clip.</i></p>

Same seed, same prompt, both quants:

<table>
<tr>
<td align="center"><b>Official comfy-int8-convrot (21.5GB)</b></td>
<td align="center"><b>NVFP4 — this repo (18.7GB)</b></td>
</tr>
<tr>
<td><img src="https://huggingface.co/BennyDaBall/LTX-2.5-22b-distilled-nvfp4-comfy/resolve/main/assets/surfer_int8.webp" alt="int8 surfer"></td>
<td><img src="https://huggingface.co/BennyDaBall/LTX-2.5-22b-distilled-nvfp4-comfy/resolve/main/assets/surfer_nvfp4.webp" alt="nvfp4 surfer"></td>
</tr>
<tr>
<td><img src="https://huggingface.co/BennyDaBall/LTX-2.5-22b-distilled-nvfp4-comfy/resolve/main/assets/disco_int8.webp" alt="int8 disco i2v"></td>
<td><img src="https://huggingface.co/BennyDaBall/LTX-2.5-22b-distilled-nvfp4-comfy/resolve/main/assets/disco_nvfp4.webp" alt="nvfp4 disco i2v"></td>
</tr>
</table>

Top row: text-to-video. Bottom row: **image-to-video** from a single keyframe — same start image, same seed, both quants. Native synced audio on all four (yes, the disco clips come with their own disco track).

## The problem

Lightricks ships an NVFP4 export of the 2.5 distilled DiT, but it's a raw TensorRT-ModelOpt export. Load it through ComfyUI's UNETLoader and you get:

```
mat1 and mat2 shapes cannot be multiplied (6820x4096 and 2048x4096)
```

That `2048x4096` is a packed 4-bit weight being read as raw bytes — exactly half the width it should be. The funny part: the file's scale tensors (`weight_scale`, `weight_scale_2`, `input_scale`) already match ComfyUI's nvfp4 spec name-for-name. The only thing missing is the per-layer `.comfy_quant` marker tensor ComfyUI uses to detect quantized layers. Official "comfy-" packs have them; the NVFP4 export doesn't.

## The fix

This repo is the official NVFP4 file with a `{"format": "nvfp4"}` marker stamped onto all **1,176 quantized layers**. No requantization, no math — the weight bytes are untouched. It just tells ComfyUI what it's looking at.

| | |
|---|---|
| Base | [Lightricks/LTX-2.5](https://huggingface.co/Lightricks/LTX-2.5) `ltx-2.5-22b-distilled-transformer-nvfp4` |
| Quant | NVFP4 (group 16, ModelOpt scales), 1,176 layers |
| Size | 18.7GB (vs 21.5GB comfy-int8-convrot) |
| Needs | ComfyUI ≥ v0.32.0, **Blackwell GPU** (RTX 50-series) for the FP4 kernels |

## Usage

Drop it in `ComfyUI/models/diffusion_models/` and pick it as `unet_name` in the stock **LTX-2.5 Text to Video / Image to Video** templates (ComfyUI ships them natively — no custom nodes). Text encoder, VAEs, and the latent upscaler come from the official repo as usual.

Lost? Grab [`ltx-2.5-nvfp4-t2v-example-workflow.json`](./ltx-2.5-nvfp4-t2v-example-workflow.json) from this repo — the official T2V template with this model pre-selected. Load it, type a prompt, hit Queue.

## Receipts

- Not pixel-identical (different quant noise), but close — same composition, same detail level, same motion beats. You'd have to flip between them to tell.
- 10s @ 1280×736 with synced audio in ~50s on an RTX 5090, including first model load.
- The 2.8GB of freed weight VRAM is real headroom against the long-duration offload cliff — the difference between finishing a clip and paging to death.

## Credits

- **Lightricks** for LTX-2.5 and the source NVFP4 export.
- Conversion + testing: [BennyDaBall_OG](https://x.com/BennyDaBall_OG) — follow me on X.

## Disclaimer

Weights are governed by the LTX-2.x Community License (included). Use at your own risk; what you generate is on you.