Skip to content
WorkflowListedExternal
Open source

REAL TIME INPAINTING WORKFLOW — v1.0

Just rolled out a real-time inpainting pipeline using SAM2 and Comfystream , running on an 4090 via RunPod. Followed this guide to get runpod template and required models.

Sourcecivitai.com/models/1553951
TaskMultiCapabilitiesImage edit · Image to imageFamilyStable DiffusionEstimated VRAMNot yet characterizedBgrade

Runtime profile

Access
External workflow
Status
Listed
Source
Civitai
Updated
Sep 7, 2026

Source description

Real-Time Inpainting Workflow

Just rolled out a real-time inpainting pipeline using SAM2 and Comfystream , running on an 4090 via RunPod. Followed this guide https://pipelines.livepeer.org/docs/technical/getting-started/install-comfystream#remote-gpu to get runpod template and required models.

Custom nodes:

SAM2 for segmentation: https://github.com/neverbiasu/ComfyUI-SAM2

Inpainting modules :

Crop Stitch: https://github.com/lquesada/ComfyUI-Inpaint-CropAndStitch

Impact Pack: https://github.com/ltdrdata/ComfyUI-Impact-Pack

I'll be sharing more real-time workflows soon— follow me on X to stay updated !

https://x.com/nieltenghu

What you need

Hardware and download-size estimates are not available yet.

Requirements

Models

9

  • CLIPLoader

    Possible

    Text encoder · Unknown

  • CLIPText/model.fp16.safetensors

    Not resolved

    Text encoder · Hugging Face · CLIPText/model.fp16.safetensors

  • control_v11f1p_sd15_depth_fp16.safetensors

    Not resolved

    ControlNet · Unknown

  • GroundingDinoModelLoader (segment anything2)

    File unverified

    Segmentation · Unknown

  • GroundingDinoSAM2Segment (segment anything2)

    File unverified

    Segmentation · Unknown

  • SAM2ModelLoader (segment anything2)

    File unverified

    Segmentation · Unknown

  • TorchCompileLoadVAE

    File unverified

    VAE · Unknown

  • VAEDecode

    Possible

    VAE · Unknown

  • VAELoader

    Possible

    VAE · Unknown

Node packs

17

  • CLIPTextEncode

    Possible

    Node pack · Unknown

  • ControlNetApplyAdvanced

    Possible

    Node pack · Unknown

  • ControlNetLoader

    Possible

    Node pack · Unknown

  • DepthAnythingTensorrt

    Found

    Node pack · Unknown

  • DifferentialDiffusion

    Not resolved

    Node pack · Unknown

  • FeatureBankAttentionProcessor

    Found

    Node pack · Unknown

  • GrowMask

    Possible

    Node pack · Unknown

  • ImageResize+

    Possible

    Node pack · Unknown

  • ImpactGaussianBlurMask

    Found

    Node pack · Unknown

  • InpaintCropImproved

    Found

    Node pack · Unknown

  • InpaintModelConditioning

    Possible

    Node pack · Unknown

  • InpaintStitchImproved

    Found

    Node pack · Unknown

  • KSampler

    Possible

    Node pack · Unknown

  • LoadImage

    Possible

    Node pack · Unknown

  • PreviewImage

    Possible

    Node pack · Unknown

  • TensorRTLoader

    Found

    Node pack · Unknown

  • TorchCompileLoadControlNet

    Found

    Node pack · Unknown

AI-generated commentary

AI-generated explanation based on source and configuration details. Suggestions are clearly labeled.

Purpose

  • Use this workflow to inpaint an input image: LoadImage names the image input, SAM2 handles segmentation, Crop Stitch and Impact Pack are the inpainting modules, and PreviewImage is the named preview output.

Flow

  • Other listed processing nodes include InpaintCropImproved, InpaintStitchImproved, KSampler, and VAEDecode.

Inputs and outputs

  • LoadImage is the named image input, and PreviewImage is the named preview output.

Setup

  • The described setup uses Comfystream on a 4090 via RunPod and links a remote-GPU guide for the RunPod template and required models.

Dependencies

  • Named node packs include ComfyUI-SAM2, ComfyUI-Inpaint-CropAndStitch, and ComfyUI-Impact-Pack.

  • Other listed node packs are ComfyUI-Depth-Anything-Tensorrt, ComfyUI-Torch-Compile, ComfyUI-Stream-Pack, and ComfyUI_TensorRT.

  • Other listed node names are CLIPLoader; CLIPTextEncode; ControlNetApplyAdvanced; ControlNetLoader; DepthAnythingTensorrt; DifferentialDiffusion; FeatureBankAttentionProcessor; GroundingDinoModelLoader (segment anything2); GroundingDinoSAM2Segment (segment anything2); GrowMask; I‌

Configuration

  • The listed base model is SD 1.5.

Cautions

  • Before use, check that your local files include CLIPText/model.fp16.safetensors and control_v11f1p_sd15_depth_fp16.safetensors.

    Suggestion · not verified

Does this need to be edited?

Sign in to send an edit request.

Sources

1 source
CivitaiPrimary
civitai.com/models/1553951

Verified Jul 19

Source excerpts

1 excerpt
NoteJul 19civitai.com/models/1553951

Source context: 142 downloads · Type Workflows · Base model SD 1.5