---
license: other
license_name: ltx-2-community-license
license_link: https://huggingface.co/Patarapoom/avatarforever-weights/blob/main/LICENSE-LTX-2.md
tags:
- avatar
- audio-to-video
- talking-head
- ltx-2
---

# AvatarForever weights: unmodified mirror

This repository is an unmodified mirror of the inference weights for
[Avatar-Forever](https://github.com/leeruibin/avatarforever) (arXiv 2608.12107).
It is laid out as a single repository so it can be pre-cached by RunPod Serverless "cached models".

Nothing here was retrained, merged, quantised or otherwise modified. The one exception is the nested
`gemma-3-12b-it-qat-q4_0-unquantized/README.md`: its `extra_gated_*` front-matter keys were removed.
The source mirror is not gated, and RunPod's cached-model UI read those keys as a gate.
Every weight, tokenizer and config file is byte-identical to its source.

| path | source | revision | bytes |
|---|---|---|---|
| `avatarforever/avatarforever-ltx-2.3-22b.safetensors` | [LetsThink/AvatarForever](https://huggingface.co/LetsThink/AvatarForever) | `622ea82482c09e7fb2d8e29cb7ff1d009c820f05` | 46,183,958,894 (sha256 `08dea67ed3d35a75da49b056ce2ca59a0e9ce538b429143fb9022c9bd5c3b145`) |
| `gemma-3-12b-it-qat-q4_0-unquantized/` | [Lightricks/gemma-3-12b-it-qat-q4_0-unquantized](https://huggingface.co/Lightricks/gemma-3-12b-it-qat-q4_0-unquantized), byte-identical to [google/gemma-3-12b-it-qat-q4_0-unquantized](https://huggingface.co/google/gemma-3-12b-it-qat-q4_0-unquantized) | `d62fe4f1995ade703b49a0f3c0d0f161237ef437` | 24,414,161,761 |

## Licences

- **AvatarForever checkpoint**: an LTX-2 derivative, distributed under the **LTX-2 Community License Agreement**
  (see [`LICENSE-LTX-2.md`](LICENSE-LTX-2.md), copied verbatim from the upstream repository).
  - Your use is subject to paragraph 4 and **Attachment A (Use Restrictions)** of that agreement, in their entirety.
  - Entities with annual revenue of USD 10,000,000 or more need a commercial agreement with Lightricks.
- **Gemma text encoder** (`gemma-3-12b-it-qat-q4_0-unquantized/`): Gemma is provided under and subject to the
  Gemma Terms of Use found at [ai.google.dev/gemma/terms](https://ai.google.dev/gemma/terms).
  - Your use is subject to the [Gemma Prohibited Use Policy](https://ai.google.dev/gemma/prohibited_use_policy).
  - See [`NOTICE`](NOTICE).

## Use

Follow the upstream inference instructions at <https://github.com/leeruibin/avatarforever> and point:

- `--distilled-checkpoint-path` at `avatarforever/avatarforever-ltx-2.3-22b.safetensors`
- `--gemma-root` at `gemma-3-12b-it-qat-q4_0-unquantized/`

The upstream code at `de4deb6a` needs `transformers` 4.x. It was verified with 4.57.6; 5.x fails while loading Gemma.