Wan2.1 InfiniteTalk LipSync (Native) — v1.5

This ComfyUI workflow generates lip-synced talking-head videos from a face image and speech audio, using the audio duration to create repeated video segments and outputting an MP4 with the original audio.

It analyzes speech duration, calculates a loop count, then generates an initial video block and extensions guided by ending frames from the previous block.

ImageBatch appends new frame segments to the accumulated video, while Save Video [Eclipse] combines the final image batch with the original audio in an MP4.

Use a starting face image and a voice clip; the audio encoder converts the clip into phonemic feature representations.

The output is an MP4 assembled from the accumulated image batch and original audio; trim_mode is set to shortest.

The workflow runs in ComfyUI and requires Eclipse 4.1.22; its listed node packs are ComfyUI_Eclipse and comfyui-easy-use.

Local setup places model files in ComfyUI subfolders for diffusion_models, text_encoders, model_patches, audio_encoders, and vae.

Listed model files include nsfw_wan_umt5-xxl_bf16_fixed.safetensors, wan/lightx2v_I2V_14B_480p_cfg_step_distill_rank64_bf16.safetensors, wan/Wan2_1-I2V-14B-720p_fp8_e4m3fn_scaled_KJ.safetensors, and Wan2_1_VAE_bf16.safetensors.

The workflow also lists wan2.1_infiniteTalk_multi_fp16.safetensors and wav2vec2-chinese-base_fp16.safetensors.

Its node-pack requirements are ComfyUI_Eclipse and comfyui-easy-use.

Choose auto-calculated loops based on audio duration or manual loops controlled by the loop_count setting.

The Settings group exposes output width and height, target frame rate, and loop_count; examples include 480p or 720p and frame rates of 24.0 or 30.0.

Verify that Wan2_1_VAE_bf16.safetensors, wan/lightx2v_I2V_14B_480p_cfg_step_distill_rank64_bf16.safetensors, and wav2vec2-chinese-base_fp16.safetensors are available before use.

Suggestion · non vérifié

Confirm Eclipse 4.1.22 and the listed node packs are installed before use.

Suggestion · non vérifié