Time-Adapter / README.md
Constellum's picture
Release TA-only step5750 checkpoint and config
176ab82
|
Raw
History Blame Contribute Delete
3.14 kB
---
license: mit
language:
- en
- ru
base_model:
- kandinskylab/Kandinsky-5.0-I2V-Lite-5s
pipeline_tag: image-to-video
tags:
- image-to-video
- video-generation
- kandinsky-5
- time-adapter
- fps-conditioning
- latent-time
widget:
- text: >-
Target FPS: 15. A close-up of a green-eyed girl in a dress stands in the
city. The girl smiles and waves. Then she turns her back and walks away.
output:
url: examples/ta_pose_15.mp4
- text: >-
Target FPS: 30. A close-up of a green-eyed girl in a dress stands in the
city. The girl smiles and waves. Then she turns her back and walks away.
output:
url: examples/ta_pose_30.mp4
- text: >-
Target FPS: 60. A close-up of a green-eyed girl in a dress stands in the
city. The girl smiles and waves. Then she turns her back and walks away.
output:
url: examples/ta_pose_60.mp4
- text: >-
Target FPS: 15. Beautiful views of the forest with fog. The wind dispels
the fog and the sun peeks out over the forest. The sun's rays illuminate a
beautiful green forest.
output:
url: examples/ta_forest_15.mp4
- text: >-
Target FPS: 30. Beautiful views of the forest with fog. The wind dispels
the fog and the sun peeks out over the forest. The sun's rays illuminate a
beautiful green forest.
output:
url: examples/ta_forest_30.mp4
- text: >-
Target FPS: 60. Beautiful views of the forest with fog. The wind dispels
the fog and the sun peeks out over the forest. The sun's rays illuminate a
beautiful green forest.
output:
url: examples/ta_forest_60.mp4
---
<div style="background-color: #f8f9fa; padding: 20px; border-radius: 10px; margin-bottom: 20px;">
<h1 style="color: #24292e; margin-top: 0;">Time Adapter for Kandinsky-5 I2V</h1>
<div style="background-color: white; padding: 15px; border-radius: 8px; margin: 15px 0; box-shadow: 0 2px 4px rgba(0,0,0,0.1);">
<h2 style="color: #24292e; margin-top: 0;">Overview</h2>
<p>This repository contains a lightweight TA-only temporal adapter checkpoint for Kandinsky-5 image-to-video generation.</p>
<p>Lower target FPS typically leads to slower perceived motion, while higher target FPS leads to faster perceived motion.</p>
<p>The released checkpoint keeps the base DiT weights unchanged and adds only the time-adapter parameters for global FPS conditioning and latent-time control.</p>
<p>Use this checkpoint with the `time_adapter` branch of <code>kandinsky-5</code> and the config <code>configs/k5_i2v_time_adapter_step5750_ta_only.yaml</code>.</p>
</div>
<div style="background-color: white; padding: 15px; border-radius: 8px; margin: 15px 0; box-shadow: 0 2px 4px rgba(0,0,0,0.1);">
<h2 style="color: #24292e; margin-top: 0;">Examples</h2>
<p>The gallery below shows TA-only examples for pose motion control and forest motion control at different target FPS values.</p>
<p>The config follows the standard `kandinsky-5` layout and expects model files under <code>weights/model</code>, <code>weights/vae</code>, <code>weights/text_encoder</code>, and <code>weights/text_encoder2</code>.</p>
</div>
</div>
<Gallery />