Here is another simple example of using the Depth control. Many steps are similar to the Edge example. The main difference is to use assets/inference_cosmos_transfer1_single_control_depth.json as the --controlnet_specs:
export CUDA_VISIBLE_DEVICES=0
export CHECKPOINT_DIR="${CHECKPOINT_DIR:=./checkpoints}"
PYTHONPATH=$(pwd) python cosmos_transfer1/diffusion/inference/transfer.py \
--checkpoint_dir $CHECKPOINT_DIR \
--video_save_folder outputs/example1_single_control_depth \
--controlnet_specs assets/inference_cosmos_transfer1_single_control_depth.json \
--offload_text_encoder_modelSame as the Edge example, the input video is a low-resolution 640 × 480 video.
example1_depth.mp4
This will generate a 960 x 704 video that preserves the 3D spatial structure and scene depth from the input video while enhancing visual quality, detail, and realism.