Spaces:

---

multimodalart

/

wan2-1-fast

like 865

Running on Zero

model_id = "Wan-AI/Wan2.1-I2V-14B-480P-Diffusers"
image_encoder = CLIPVisionModel.from_pretrained(
    model_id, subfolder="image_encoder", torch_dtype=torch.float32
)
vae = AutoencoderKLWan.from_pretrained(model_id, subfolder="vae", torch_dtype=torch.float32)
pipe = WanImageToVideoPipeline.from_pretrained(
    model_id, vae=vae, image_encoder=image_encoder, torch_dtype=torch.bfloat16
)
pipe.to("cuda")


 # (OPTION1) pipe.enable_sequential_cpu_offload()
 # (OPTION2) pipe.enable_model_cpu_offload()
 # (OPTION3) none. no cpu offloading.

print("Loading LoRA...")
causvid_path = hf_hub_download(repo_id="Kijai/WanVideo_comfy", filename="Wan21_CausVid_14B_T2V_lora_rank32.safetensors")
pipe.load_lora_weights(causvid_path, adapter_name="causvid_lora")
pipe.set_adapters(["causvid_lora"], adapter_weights=[0.95])
pipe.fuse_lora()
print("LoRA loaded")

Option 1

Option 2

Option 3

Using the latest Diffusers version from git, latest Pytorch nightly + CU128.
RTX 5090 32GB VRAM
64GB General RAM

Using the new pipe.fuse_lora() method recommended to me for using LoRA's with WanImageToVideoPipeline.
I have other LoRA's working for me though. I'm just leaving this here in case anybody else runs into any of these issues as well.