Wan2.1 I2v - 720p 14b Fp16.safetensors [upd]

: Place wan2.1_i2v_720p_14B_fp16.safetensors in ComfyUI/models/diffusion_models/ .

Loading this FP16 model requires ~28GB VRAM. If you have less, use the fp8 or GGUF quants instead. wan2.1 i2v 720p 14b fp16.safetensors

You will also need the text encoder (e.g., umt5-xxl-enc-bf16.safetensors ), VAE (e.g., Wan2_1_VAE_bf16.safetensors ), and CLIP models. : Place wan2

: Generally exceeds the capacity of standard consumer GPUs (like the RTX 4090/5090) when used alongside high-resolution text encoders and VAEs in a single workflow. Recommendation : Many users opt for FP8 or GGUF (quantized) versions to fit the model into 24GB VRAM. Performance wan2.1 i2v 720p 14b fp16.safetensors

NVIDIA RTX 4090, RTX 3090, or RTX 4080. Using tools like ComfyUI with low-VRAM flags allows the model to offload components to system RAM during processing steps. How to Integrate into Your Workflow