[Flux] gguf PuLID with Inpaint — v1.0

[Flux] gguf PuLID with Inpaint is for using an image and mask with PuLID inpainting to produce a saved image.

It resizes the image, encodes and decodes it with a VAE, prepares and previews the mask, guides and samples the result, and saves the image.

PuLID processing uses ApplyPulidFlux with PulidFluxModelLoader, PulidFluxInsightFaceLoader, and PulidFluxEvaClipLoader.

The meaningful inputs are an image and mask; LoadImage, ImageResize+, VAEEncode, SetLatentNoiseMask, and MaskPreview+ prepare them.

VAEDecode and SaveImage produce a saved image.

The description says you can run the workflow without setting up a ComfyUI environment, and that it is free.

The listing names Flux.1 D as the base model and Nouvo Ai as the publisher.

The model filenames are ae.safetensors, clip_l.safetensors, flux1-dev-Q8_0.gguf, pulid_flux_v0.9.0.safetensors, and t5-v1_1-xxl-encoder-Q8_0.gguf.

The listed node types include ApplyPulidFlux, BasicGuider, BasicScheduler, CLIPTextEncode, DualCLIPLoaderGGUF, FluxGuidance, ImageResize+, ImpactGaussianBlurMask, KSamplerSelect, LoadImage, and MaskPreview+.

Other listed node types are PulidFluxEvaClipLoader, PulidFluxInsightFaceLoader, PulidFluxModelLoader, RandomNoise, SamplerCustomAdvanced, SaveImage, SetLatentNoiseMask, UnetLoaderGGUF, VAEDecode, VAEEncode, and VAELoader.

To adjust denoise, try a BasicScheduler value between 0.5 and 1.

Vorschlag · nicht geprüft

Before using it, verify that t5-v1_1-xxl-encoder-Q8_0.gguf and pulid_flux_v0.9.0.safetensors are available under those exact filenames.

Vorschlag · nicht geprüft

Also verify that your setup provides the listed node types.

Vorschlag · nicht geprüft