jhj Wan Video Workflows — v1.0

Use the jhj Wan Video Workflows to generate video through separate T2V and I2V paths, with text passed through CLIPTextEncode or an image loaded through LoadImage; the generated video can then be upscaled.

The flow generates video at a smaller resolution before upscaling it for VRAM limits.

Separate I2V and T2V workflows are provided.

Use an image with the I2V path or text with the T2V path; both paths are described as producing video.

This is version v1.0, published by jhj0517.

Named model files include: 4x_foolhardy_Remacri.pth; clip_vision_h.safetensors; umt5_xxl_fp8_e4m3fn_scaled.safetensors; wan_2.1_vae.safetensors; wan2.1-i2v-14b-720p-Q4_K_S.gguf; wan2.1-t2v-14b-Q4_K_S.gguf.

Named components include WanImageToVideo, EmptyHunyuanLatentVideo, LoadImage, KSampler, CLIPTextEncode, CLIPVisionEncode, and CLIPVisionLoader.

They also include CLIPLoader, VAELoader, VAEDecode, ModelSamplingSD3, LoaderGGUF, UpscaleModelLoader, ImageScaleBy, ImageUpscaleWithModel, VHS_VideoCombine, easy loraStack, and CR Apply LoRA Stack.

Choose a quantized GGUF variant that matches your GPU: Wan2.1-I2V-14B-720P-gguf, Wan2.1-I2V-14B-480P-gguf, or Wan2.1-T2V-14B-gguf.

Suggestion · not verified

No license information is listed for this workflow.

If you encounter out-of-memory during the upscale nodes, bypass those nodes and upscale separately.

Suggestion · not verified