Thanks for the amazing work and the impressive performance!
I have a question about its input: can we provide multiple reference images for the identical object when running customization? If so, e.g., do I just simply input two images of one object into the --ref_image argument? For instance:
torchrun \
--node_rank=0 \
--nnodes=1 \
--rdzv_endpoint=127.0.0.1:23468 \
--nproc_per_node=8 generate.py --task s2v-14B --size 1280*720 --frame_num 121 --sample_fps 24 --ckpt_dir ./hf_model/Wan2.1-T2V-1.3B --phantom_ckpt ./hf_model/Phantom_hf_model --ref_image "./reference/woman_1.png,./reference/woman_2.png" --dit_fsdp --t5_fsdp --ulysses_size 8 --ring_size 1 --prompt "A young Chinese woman is standing and introducing the UNNY CLUB sun cream. She is wearing a mint - green dress and a white sun protection shawl. The background is the terrace of a seaside vacation home. The wooden floor is clean and tidy, and a blue sea can be seen in the distance. There is a beach chair and a sunshade on the right. The overall atmosphere is relaxed and cozy."
# You can generate 1280*720 video by setting "--size 1280*720"
Again, thanks for your effort of developing such impressive work!
Thanks for the amazing work and the impressive performance!
I have a question about its input: can we provide multiple reference images for the identical object when running customization? If so, e.g., do I just simply input two images of one object into the --ref_image argument? For instance:
torchrun \ --node_rank=0 \ --nnodes=1 \ --rdzv_endpoint=127.0.0.1:23468 \ --nproc_per_node=8 generate.py --task s2v-14B --size 1280*720 --frame_num 121 --sample_fps 24 --ckpt_dir ./hf_model/Wan2.1-T2V-1.3B --phantom_ckpt ./hf_model/Phantom_hf_model --ref_image "./reference/woman_1.png,./reference/woman_2.png" --dit_fsdp --t5_fsdp --ulysses_size 8 --ring_size 1 --prompt "A young Chinese woman is standing and introducing the UNNY CLUB sun cream. She is wearing a mint - green dress and a white sun protection shawl. The background is the terrace of a seaside vacation home. The wooden floor is clean and tidy, and a blue sea can be seen in the distance. There is a beach chair and a sunshade on the right. The overall atmosphere is relaxed and cozy." # You can generate 1280*720 video by setting "--size 1280*720"Again, thanks for your effort of developing such impressive work!