Model card: teaser, keyboard actions, compact layout

README.md

@@ -6,51 +6,92 @@ tags:
   - world-model
   - camera-control
   - video-generation
+  - long-video
 ---
 
-# Memorizon (4-step)
+<div align="center">
 
-**Memorizon: Training World Models Beyond Their Context Window**
-Tingting Liao, Xuezhi Liang, Hao Li, Guangyi Liu · IFM, MBZUAI
+<h2>Memorizon</h2>
+<h3>Training World Models Beyond Their Context Window</h3>
 
-[Code](https://github.com/TingtingLiao/memorizon) · [Project page](https://tingtingliao.github.io/memorizon/)
+**Tingting Liao · Xuezhi Liang · Hao Li · Guangyi Liu**
+<br>IFM, MBZUAI
 
-Memorizon turns one image and a camera trajectory into a long video, generated one
-second at a time, that stays consistent when the camera returns to places it has
-already seen. Each chunk retrieves its most relevant past frames by camera
-co-visibility. This checkpoint is the 4-step distilled model (Self-Forcing + DMD,
-teacher guidance distilled in): use 4 denoising steps and no classifier-free
-guidance. It is built on Wan2.2-TI2V-5B and outputs 864x480 video at 16 fps.
+<a href="https://github.com/TingtingLiao/memorizon"><img src="https://img.shields.io/badge/GitHub-code-black?logo=github&labelColor=white"></a>&ensp;
+<a href="https://tingtingliao.github.io/memorizon/"><img src="https://img.shields.io/badge/project-page-green?logo=googlechrome&logoColor=green&labelColor=white"></a>&ensp;
+<a href=""><img src="https://img.shields.io/badge/arXiv-coming%20soon-red?logo=arxiv&logoColor=darkred&labelColor=white"></a>
 
-## Usage
+<br>
+
+<img src="roundtrip.gif" width="85%">
+
+<sub><code>w*16 l*24 w*16 l*24</code> — walk ahead, turn around, walk back, turn around: the start view comes back. 2× speed.</sub>
+
+</div>
+
+---
+
+## ✨ This checkpoint
+
+| | |
+|---|---|
+| **Model** | Memorizon, 4-step distilled (Self-Forcing + DMD, CFG distilled in) |
+| **Base** | Wan2.2-TI2V-5B |
+| **Input** | one image + keyboard actions or a camera trajectory |
+| **Output** | 864×480, 16 fps, generated one second at a time |
+| **Sampling** | 4 steps, no CFG (read from the config) |
+
+## 🚀 Usage
 
 ```bash
-git clone https://github.com/TingtingLiao/memorizon.git && cd memorizon
-pip install -e .
-python scripts/generate.py --checkpoint Luffuly/memorizon \
-    --image your_photo.jpg --trajectory assets/trajectories/walk_00.npy --seconds 60 --output out.mp4
+git clone https://github.com/TingtingLiao/memorizon.git && cd memorizon && pip install -e .
+
+python scripts/generate.py --checkpoint Luffuly/memorizon --image photo.jpg \
+    --actions "w*16 l*24 w*16 l*24" --output out.mp4
 ```
 
 ```python
-import numpy as np
-from memorizon import MemorizonPipeline, save_video
+from memorizon import MemorizonPipeline, actions_to_c2w, save_video
 
 pipe = MemorizonPipeline.from_pretrained("Luffuly/memorizon")
-video = pipe("your_photo.jpg", np.load("assets/trajectories/walk_00.npy"),
-             num_chunks=60)   # 4 steps, no CFG; prompt=None captions the image
+video = pipe("photo.jpg", actions_to_c2w("w*16 l*24 w*16 l*24"))
 save_video(video, "out.mp4")
 ```
 
-Prompts must follow the training caption format (a perspective prefix such as
-"First-person perspective — character not visible." and one descriptive paragraph).
-Without a prompt the image is captioned automatically with Qwen3-VL-8B in that
-format (see `memorizon/caption.py`).
+### 🎮 Keyboard actions — one per 0.25 s
+
+| Key | Action | Key | Action |
+|:---:|:---|:---:|:---|
+| `w` `s` | forward / backward · 0.25 m | `j` `l` | turn left / right · 7.5° |
+| `a` `d` | left / right · 0.25 m | `i` `k` | look up / down · 7.5° |
+| `.` | stay | `*N` | repeat N times |
+
+<details>
+<summary><b>Prompts</b> — captioned automatically</summary>
+
+<br>
+
+Without a prompt, the image is captioned by Qwen3-VL-8B in the training format
+(`memorizon/caption.py`). Custom prompts should follow it — a perspective prefix,
+then one paragraph:
+
+> First-person perspective — character not visible. A winding paved path curves
+> beneath a canopy of vibrant pink cherry blossoms, …
+
+</details>
+
+<details>
+<summary><b>Camera trajectories</b></summary>
+
+<br>
+
+Instead of actions, pass `--trajectory` a `[T, 4, 4]` array of camera-to-world poses,
+`T = 1 + 4 × seconds`, in the first frame's coordinates (OpenCV axes), translations
+in metres / 4.
 
-The camera trajectory is a `[T, 4, 4]` array of camera-to-world poses, one per latent
-frame (4 per second), in the first frame's coordinates (OpenCV axes), with
-translations in metres / 4. See the code repository for details.
+</details>
 
-## Citation
+## 📖 Citation
 
 ```bibtex
 @article{memorizon2026,