Wan2.1 Vace T2I_Inpainting&out_gguf_8g_LowVram_feet&hand_fix_easy

Use Wan2.1 Vace T2I_Inpainting&out_gguf_8g_LowVram_feet&hand_fix_easy to edit a single image with inpainting or outpainting: provide an image and mask, then produce an edited image.

It includes two workflow files: Wan2.1_Vace_T2I_inpainting_outpainting_gguf_LowVram_easy_V1.json and Wan2.1_Vace_T2I_inpainting_outpainting_gguf_LowVram_easy_V1 (add_Nag).json.

The input is one image and a mask that identifies the area to change or extend.

The graph includes Image Comparer (rgthree), MaskPreview+, and SaveImage.

It is listed as a workflow from SamLiu using the Wan Video 14B i2v 480p base model.

The listed model files are umt5_xxl_fp8_e4m3fn_scaled.safetensors, wan_2.1_vae.safetensors, and Wan2.1_14B_VACE-Q4_0.gguf.

Image and mask handling names include LoadImage, ImagePadForOutpaint, ImageCompositeMasked, MaskToImage, ImageResizeKJv2, ImageGenResolutionFromImage, MaskPreview+, ResizeMask, INPAINT_MaskedFill, VAELoader, VAEDecode, and SaveImage.

Other named components include Power Lora Loader (rgthree), CLIPLoader, CFGZeroStar, CLIPTextEncode, SkipLayerGuidanceDiT, Image Comparer (rgthree), ModelSamplingSD3, KSamplerWithNAG, MarkdownNote, UnetLoaderGGUF, UNetTemporalAttentionMultiply, and WanVaceToVideo.

Use width and height values divisible by 32; the listing says other values cause an error.

For generation steps, the listing suggests 4 for a preview, 8 or 12 for “Barely,” and at least 16 for “Basically the best it can do.”

Suggestion · non vérifié

Before use, verify that the listed model files and named components are available in your setup.

Suggestion · non vérifié

The listing does not specify a license.