We provide visualizations of the generated manipulation videos and real-world robot deployments, each stored in separate folders named `Generated_Video` and `Real_Robot_Deployment`. 

- **Generated_Video**: This directory contains manipulation videos generated by *DreamTransfer* under diverse text-controlled appearance variations. Each subfolder corresponds to a task (`clean_desk`, `fold_cloth`, `throw_bottle`), containing 6 generated video clips and a `prompt.json` file that records the text prompts used for generation. Each generated video is a composite frame consisting of three rows: the top row shows the demonstration video from real-world and simulated data, the middle row displays the corresponding depth video, and the bottom row presents the *DreamTransfer*-generated video. These videos demonstrate *DreamTransfer*'s ability to preserve 3D structure and geometric plausibility while editing foreground objects, backgrounds, and lighting conditions.

- **Real_Robot_Deployment**: This directory contains videos of the trained policy executing tasks on a real robotic platform using *AdaMix* for training. Each subfolder corresponds to a task, containing 4 demonstrations under zero-shot visual conditions with diverse appearance variations. These videos illustrate the robustness and generalization of the policy learned with *AdaMix* when deployed in real-world scenarios with novel appearances.

All videos are provided in MP4 format. For details on text prompts used in video generation, please refer to the `prompt.json` files in each task subfolder.