Extensions/ComfyUI-NetflixVoid
ComfyUI Extension

ComfyUI-NetflixVoid

Custom ComfyUI nodes for the VOID workflow with five stages: black mask, VLM prep, grey mask, quadmask, and modular PQ5 inference. (Description by CC)

By jaskirat05·Created 5 months ago·Updated 5 months ago· 0
jaskirat05/ComfyUI-NetflixVoid
Nodes15
On cloudLocal install
CategoryVOID, VOID/PQ5
Stars0
Updated5 months ago

Nodes (15)

VOID Build Grey Mask

The SAM3 node that decides what falls, what stays, and what disappears

VOID
VOID Combine Quadmask

Why object removal needs four mask levels, not one

VOID
VOID Export Black Mask

The boring node that makes the whole VOID workflow agree on what's being removed

VOID
VOID Gemma 4 E2B Video Prompt

The pack's built-in VLM, heavy but local

VOID
VOID Load SAM3 Model

The grey-mask stage's auto-downloading SAM3 loader

VOID
VOID Parse VLM Analysis

Turning your VLM's rambling into the JSON the rest of the VOID graph demands

VOID
VOID PQ5 Decode Video

Latents back into frames, trimmed to your real video length

VOID/PQ5
VOID PQ5 Encode Prompt

Turn your edit intent into T5 embeds the VOID sampler can use

VOID/PQ5
VOID PQ5 Encode Quadmask

Snap your 4-level mask to the shape the VOID model demands

VOID/PQ5
VOID PQ5 Encode Video

Prep your footage for the VOID model (and keep track of the original frame count)

VOID/PQ5
VOID PQ5 Load Model

The loader with no fallbacks (and why that's on you)

VOID/PQ5
VOID PQ5 Sampler

The node that actually regenerates your video

VOID/PQ5
VOID PQ5 Settings

One settings object, shared across your whole PQ5 graph

VOID/PQ5
VOID PQ5 Unload Cache

The memory reset button for the VOID pipeline

VOID/PQ5
VOID Prepare VLM Analysis

The node that turns 'what breaks if I delete this?' into a question a vision model can answer

VOID
Readme

ComfyUI-NetflixVoid

Custom ComfyUI nodes for the VOID workflow:

  1. Stage 1: black mask
  2. Stage 2: VLM prep + parse
  3. Stage 3: grey mask
  4. Stage 4: quadmask
  5. Stage 5: modular PQ5 inference

Included nodes

  • VOID Load SAM3 Model
  • VOID Export Black Mask
  • VOID Prepare VLM Analysis
  • VOID Parse VLM Analysis
  • VOID Build Grey Mask
  • VOID Combine Quadmask
  • VOID Gemma 4 E2B Video Prompt
  • VOID PQ5 Load Model
  • VOID PQ5 Settings
  • VOID PQ5 Encode Prompt
  • VOID PQ5 Encode Video
  • VOID PQ5 Encode Quadmask
  • VOID PQ5 Sampler
  • VOID PQ5 Decode Video
  • VOID PQ5 Unload Cache

PQ5 assets and models

This repo now uses a modular PQ5 layout.

Required locations:

  • Transformer checkpoint: ComfyUI/models/checkpoints (select in VOID PQ5 Load Model)
  • VAE checkpoint: ComfyUI/models/vae (select in VOID PQ5 Load Model)
  • Text encoder folder: ComfyUI/models/text_encoders/void
  • Repo-local PQ5 configs/tokenizer/scheduler: ComfyUI-NetflixVoid/pq5_assets

Recommended modular wiring (PQ5)

  1. VOID PQ5 Load Model -> model
  2. VOID PQ5 Encode Prompt with model + prompt text
  3. VOID PQ5 Encode Video with model + input video
  4. VOID PQ5 Encode Quadmask with quadmask video
  5. VOID PQ5 Sampler with model, prompt embeds, encoded video, encoded quadmask, and settings
  6. VOID PQ5 Decode Video with model, sampler latents, and original_frame_count

For pass 2:

  • Load a second model using pass-2 checkpoint in VOID PQ5 Load Model
  • Feed pass-1 output video into VOID PQ5 Encode Video
  • Reuse/refine quadmask and run sampler+decode again

Notes

  • VOID PQ5 Load Model has no fallback for missing checkpoint/VAE/text-encoder paths.
  • Use VOID PQ5 Unload Cache to clear cached PQ5 pipelines and free memory.
  • Keep grid_rows/grid_cols from VOID Prepare VLM Analysis connected to VOID Build Grey Mask.