DanJZY commited on
Commit
6d29bb6
·
verified ·
1 Parent(s): 9e1f36d

Document public AudioLift checkpoint downloads

Browse files

## Summary
Link the public inference repository and document model downloads without login. Preserve checkpoint bytes and component licenses.

## Files Created / Modified
- README.md (Modified)

Files changed (1) hide show
  1. README.md +8 -5
README.md CHANGED
@@ -17,7 +17,7 @@ tags:
17
  **Audio-Driven Full-Head Gaussian Avatar via Multi-view Video Generation**
18
 
19
  This repository provides the two fine-tuned models used by AudioLift.
20
- The weights are currently private for release review.
21
 
22
  | Model | File | Purpose |
23
  | --- | --- | --- |
@@ -30,18 +30,21 @@ have changed. File checksums are recorded in [SHA256SUMS](SHA256SUMS).
30
  ## Usage
31
 
32
  Use the AudioLift inference entrypoints in the
33
- [code review branch](https://github.com/ZhuoyuanJiang/StableAvatar-Multiview/tree/release/AudioLift).
34
  The entrypoints download missing models to the local `checkpoints/` directory
35
  and verify the fine-tuned checkpoint checksums before loading them.
36
- During private review, authenticate with a Hugging Face account that has access.
 
 
 
37
 
38
  AudioLift-mvdiffusion also requires the Wan and wav2vec2 assets from
39
  [StableAvatar](https://huggingface.co/FrancisRing/StableAvatar). These are downloaded
40
  from the upstream model repository. AudioLift-gaussianlifter requires the camera
41
  and model configuration files supplied with the AudioLift code.
42
 
43
- These are inference checkpoints. The four-view Stage-1 training initialization
44
- is a separate release asset and is not included here yet.
45
 
46
  ## Licenses and acknowledgements
47
 
 
17
  **Audio-Driven Full-Head Gaussian Avatar via Multi-view Video Generation**
18
 
19
  This repository provides the two fine-tuned models used by AudioLift.
20
+ The weights are publicly downloadable without a Hugging Face login.
21
 
22
  | Model | File | Purpose |
23
  | --- | --- | --- |
 
30
  ## Usage
31
 
32
  Use the AudioLift inference entrypoints in the
33
+ [AudioLift repository](https://github.com/ZhuoyuanJiang/AudioLift).
34
  The entrypoints download missing models to the local `checkpoints/` directory
35
  and verify the fine-tuned checkpoint checksums before loading them.
36
+ To download both models and their required base assets before inference, run
37
+ `python scripts/download_models.py` from an installed AudioLift environment.
38
+ See the [installation guide](https://github.com/ZhuoyuanJiang/AudioLift/blob/main/docs/installation.md)
39
+ for environment setup and validation status.
40
 
41
  AudioLift-mvdiffusion also requires the Wan and wav2vec2 assets from
42
  [StableAvatar](https://huggingface.co/FrancisRing/StableAvatar). These are downloaded
43
  from the upstream model repository. AudioLift-gaussianlifter requires the camera
44
  and model configuration files supplied with the AudioLift code.
45
 
46
+ These are inference checkpoints. Training code is not included in the public
47
+ AudioLift code release.
48
 
49
  ## Licenses and acknowledgements
50