diff --git a/README.md b/README.md
index 9814170..7b728a9 100644
--- a/README.md
+++ b/README.md
@@ -1,14 +1,12 @@
 # Stable Virtual Camera
 
 <a href="https://stable-virtual-camera.github.io"><img src="https://img.shields.io/badge/%F0%9F%8F%A0%20Project%20Page-gray.svg"></a>
-<a href="https://stable-virtual-camera.github.io/pdf/paper.pdf"><img src="https://img.shields.io/badge/%F0%9F%93%84%20Paper-gray.svg"></a>
+<a href="http://arxiv.org/abs/2503.14489"><img src="https://img.shields.io/badge/%F0%9F%93%84%20arXiv-2503.14489-B31B1B.svg"></a>
 <a href="https://stability.ai/news/introducing-stable-virtual-camera-multi-view-video-generation-with-3d-camera-control"><img src="https://img.shields.io/badge/%F0%9F%93%83%20Blog-Stability%20AI-orange.svg"></a>
 <a href="https://huggingface.co/stabilityai/stable-virtual-camera"><img src="https://img.shields.io/badge/%F0%9F%A4%97%20Model_Card-Huggingface-orange"></a>
 <a href="https://huggingface.co/spaces/stabilityai/stable-virtual-camera"><img src="https://img.shields.io/badge/%F0%9F%9A%80%20Gradio%20Demo-Huggingface-orange"></a>
 <a href="https://www.youtube.com/channel/UCLLlVDcS7nNenT_zzO3OPxQ"><img src="https://img.shields.io/badge/%F0%9F%8E%AC%20Video-YouTube-orange"></a>
 
-<!-- <a href="https://arxiv.org/abs/0000.00000"><img src="https://img.shields.io/badge/%F0%9F%93%84%20arXiv-2408.00653-B31B1B.svg"></a> -->
-
 `Stable Virtual Camera (Seva)` is a 1.3B generalist diffusion model for Novel View Synthesis (NVS), generating 3D consistent novel views of a scene, given any number of input views and target cameras.
 
 # :tada: News
@@ -49,7 +47,7 @@ python demo.py --data_path <data_path> [additional arguments]
 
 For a more detailed guide, follow [CLI_USAGE.md](docs/CLI_USAGE.md).
 
-For users interested in benchmarking NVS models using command lines, check [`benchmark`](benchmark/) containing the details about scenes, splits, and input/target views we reported in the <a href="https://stable-virtual-camera.github.io/pdf/paper.pdf">paper</a>.
+For users interested in benchmarking NVS models using command lines, check [`benchmark`](benchmark/) containing the details about scenes, splits, and input/target views we reported in the <a href="http://arxiv.org/abs/2503.14489">paper</a>.
 
 # :books: Citing
 
diff --git a/demo_gr.py b/demo_gr.py
index dd21957..0936105 100644
--- a/demo_gr.py
+++ b/demo_gr.py
@@ -811,7 +811,7 @@ def get_preamble():
 # Stable Virtual Camera
 <span style="display: flex; flex-wrap: wrap; gap: 5px;">
     <a href="https://stable-virtual-camera.github.io"><img src="https://img.shields.io/badge/%F0%9F%8F%A0%20Project%20Page-gray.svg"></a>
-    <a href="https://stable-virtual-camera.github.io/pdf/paper.pdf"><img src="https://img.shields.io/badge/%F0%9F%93%84%20Paper-gray.svg"></a>
+    <a href="http://arxiv.org/abs/2503.14489"><img src="https://img.shields.io/badge/%F0%9F%93%84%20arXiv-2503.14489-B31B1B.svg"></a>
     <a href="https://stability.ai/news/introducing-stable-virtual-camera-multi-view-video-generation-with-3d-camera-control"><img src="https://img.shields.io/badge/%F0%9F%93%83%20Blog-Stability%20AI-orange.svg"></a>
     <a href="https://huggingface.co/stabilityai/stable-virtual-camera"><img src="https://img.shields.io/badge/%F0%9F%A4%97%20Model_Card-Huggingface-orange"></a>
     <a href="https://huggingface.co/spaces/stabilityai/stable-virtual-camera"><img src="https://img.shields.io/badge/%F0%9F%9A%80%20Gradio%20Demo-Huggingface-orange"></a>
diff --git a/docs/CLI_USAGE.md b/docs/CLI_USAGE.md
index d1295e2..7894c7b 100644
--- a/docs/CLI_USAGE.md
+++ b/docs/CLI_USAGE.md
@@ -55,7 +55,7 @@ We provide <a href="https://github.com/Stability-AI/stable-virtual-camera/releas
 
 ### Recommended Usage
 
-- `img2img` and `img2vid` are recommended to be used for evaluation and benchmarking. These two tasks are used for the quantitative evalution throught the <a href="https://stable-virtual-camera.github.io/pdf/paper.pdf">paper</a>. The data is converted from academic datasets so the groundtruth target views are available for metric computation. Check the [`benchmark`](../benchmark/) folder for detailed splits we organize to benchmark different NVS models.
+- `img2img` and `img2vid` are recommended to be used for evaluation and benchmarking. These two tasks are used for the quantitative evalution throught the <a href="http://arxiv.org/abs/2503.14489">paper</a>. The data is converted from academic datasets so the groundtruth target views are available for metric computation. Check the [`benchmark`](../benchmark/) folder for detailed splits we organize to benchmark different NVS models.
 - `img2vid` requries both the input and target views to be sorted, which is usually not guaranteed in general usage.
 - `img2trajvid_s-prob` is for general usage but only for single-view regime and fixed preset camera control.
 - `img2trajvid` is the task designed for general usage since it does not need the ordering of the input views. This is the task used in the gradio demo.