Wan2.2 i2v gguf + lightx2v lora — i2v_v1

Wan2.2 i2v gguf + lightx2v lora is a four-step workflow for turning an image into a 33-frame video.

It uses image loading, Wan image-to-video processing, VAE decoding, and video combination steps.

Use an image as input; the described output is a 33-frame video.

Lightx2v Lora must be installed to increase speed.

The author describes using an RX 6800 XT graphics card and reports about 250 seconds for 33 frames after the models are fully loaded.

The listed model downloads are Wan2.2-I2V-A14B-HighNoise-Q3_K_S.gguf, Wan2.2-I2V-A14B-LowNoise-Q3_K_S.gguf, umt5_xxl_fp8_e4m3fn_scaled, wan_2.1_vae, and Wan21_T2V_14B_lightx2v_cfg_step_distill_lora_rank32.

The workflow names these model files: umt5_xxl_fp8_e4m3fn_scaled.safetensors, wan2.2_i2v_high_noise_14B_Q3_K_S.gguf, wan2.2_i2v_low_noise_14B_Q3_K_S.gguf, and wan_2.1_vae.safetensors.

The required node packs are gguf, rgthree-comfy, and comfyui-videohelpersuite.

The listed base model is Wan Video 14B i2v 480p.

The author says they use the Lightx2v version for Wan2.1 instead of the Wan2.2 version because they consider its results better.

The named node-pack versions are gguf 2.3.2, rgthree-comfy 1.0.2507112302, and comfyui-videohelpersuite 1.6.1.

Before using it, verify that these exact files are available: umt5_xxl_fp8_e4m3fn_scaled.safetensors, wan2.2_i2v_high_noise_14B_Q3_K_S.gguf, wan2.2_i2v_low_noise_14B_Q3_K_S.gguf, and wan_2.1_vae.safetensors.

Suggestion · not verified

A license is not listed for this workflow.