diff --git a/README.md b/README.md
index 617257414..1e9508550 100644
--- a/README.md
+++ b/README.md
@@ -33,7 +33,7 @@ Large Model for All. See our [vision](https://github.com/OptimalScale/LMFlow#vis
 
 
 ## Latest News
-* [2023-07-23] :rocket: [LMFlow multimodal chatbot](https://github.com/OptimalScale/LMFlow/blob/main/scripts/run_vis_chatbot_gradio_minigpt4.sh) is now available! Support multimodal inputs of images and texts ([Online Demo](https://da57dba5265dc44ec7.gradio.live)) :rocket: ![image](https://github.com/OptimalScale/LMFlow/blob/rpan-vision-encoder/assets/multimodal-chatbot-demo.gif)
+* [2023-07-23] :rocket: [LMFlow multimodal chatbot](https://github.com/OptimalScale/LMFlow/blob/main/scripts/run_vis_chatbot_gradio_minigpt4.sh) is now available! Support multimodal inputs of images and texts :rocket: ![image](https://github.com/OptimalScale/LMFlow/blob/rpan-vision-encoder/assets/multimodal-chatbot-demo.gif)
 * [2023-06-22]  [LMFlow paper](https://arxiv.org/abs/2306.12420) is out! Check out our implementation details at https://arxiv.org/abs/2306.12420
 * [2023-06-16] Our finetuned Robin-33B-V2 scored an impressive 64.1 on the Huggingface LLM leaderboard in our offline evaluation, outperforming major open-source LLMs! All checkpoints (7B, 13B, 33B, and 65B) are [released](https://huggingface.co/OptimalScale)! Checkout the performance [here](https://medium.com/@hkust.ml/robin-v2-launches-achieves-unparalleled-performance-on-openllm-4f6886e822c1).
 * [2023-06-07] LMFlow is now officially available on PyPI! Install it with `pip install lmflow-finetune`!