comfy_extras/

The standard library of "extra" built-in nodes — everything that isn't core enough to live in nodes.py but ships out of the box. ~27,000 lines spread over ~100 modules, organized by feature/model family.

Layout

comfy_extras/
├── chainner_models/                 # Vendored ESRGAN/upscaler model code (chaiNNer port)
├── frame_interpolation_models/      # Interpolation backbones (RIFE etc.)
├── nodes_<feature>.py               # One module per node family

File naming convention

Each comfy_extras/nodes_<topic>.py module is self-contained: it imports comfy/ and comfy_api/, defines its node classes, and exposes either:

• V1 style — a NODE_CLASS_MAPPINGS dict at module scope, or
• V3 style — a comfy_entrypoint() async function returning a ComfyExtension.

The loader in nodes.init_extra_nodes (see nodes.py) walks the directory and merges what it finds.

How the modules cluster

A rough taxonomy:

Per-architecture model nodes

One module per major model family:

File	Models covered
nodes_sd3.py	SD3 / SD3.5
nodes_flux.py	Flux (dev / schnell / Kontext / Krea)
nodes_hunyuan.py	Hunyuan DiT
nodes_hunyuan3d.py	Hunyuan3D 2.0/2.1
nodes_lumina2.py	Lumina 2
nodes_hidream.py	HiDream / HiDream E1.1
nodes_pixart.py	Pixart Alpha / Sigma
nodes_qwen.py	Qwen Image / Edit
nodes_lt.py, nodes_lt_audio.py, nodes_lt_upsampler.py	LTX-Video
nodes_mochi.py	Mochi (Genmo)
nodes_wan.py, nodes_wanmove.py	Wan 2.1/2.2
nodes_ace.py	ACE Step (audio)
nodes_cosmos.py	Cosmos
nodes_kandinsky5.py	Kandinsky 5
nodes_chroma_radiance.py	Chroma Radiance
nodes_zimage.py	Z Image
nodes_video_model.py	Stable Video Diffusion
nodes_stable_cascade.py	Stable Cascade
nodes_stable3d.py	Stable Zero123-style 3D
nodes_lotus.py	Lotus depth/normals

Sampling and conditioning extras

File	What it adds
nodes_custom_sampler.py	"Custom Sampler" composable graph (largest extras file: 42 KB)
nodes_advanced_samplers.py	Heun++, restart, etc.
nodes_align_your_steps.py	AYS schedules
nodes_optimalsteps.py	Optimal-steps schedule
nodes_gits.py	GITS noise schedule (large)
nodes_eps.py	Epsilon prediction conversions
nodes_perpneg.py	Perpendicular negative prompts
nodes_pag.py, nodes_sag.py	Perturbed / Self-Attention Guidance
nodes_apg.py	Adaptive Projected Guidance
nodes_slg.py, nodes_tcfg.py, nodes_nag.py, nodes_cfg.py	Various CFG variants
nodes_freelunch.py, nodes_fresca.py, nodes_mahiro.py, nodes_hypertile.py	Inference tweaks
nodes_easycache.py	Inference caching for diffusion (FBCache-like)
nodes_torch_compile.py	torch.compile wrapper
nodes_attention_multiply.py	Attention scale tricks
nodes_differential_diffusion.py	Differential Diffusion
nodes_context_windows.py	Sliding window context (mainly for video)
nodes_rope.py, nodes_camera_trajectory.py	RoPE and video camera helpers

Conditioning, loaders, hooks

File	What it adds
nodes_cond.py	Conditioning combinators
nodes_clip_sdxl.py, nodes_textgen.py	SDXL CLIP and text generation
nodes_hooks.py	Hook framework UI nodes
nodes_controlnet.py	ControlNet apply nodes
nodes_model_advanced.py, nodes_model_downscale.py, nodes_model_merging.py, nodes_model_merging_model_specific.py, nodes_model_patch.py	Model surgery
nodes_lora_debug.py, nodes_lora_extract.py	LoRA tooling
nodes_hypernetwork.py	Hypernetwork loader
nodes_photomaker.py	PhotoMaker
nodes_ip2p.py	InstructPix2Pix
nodes_gligen.py (in core), nodes_edit_model.py	GLIGEN, edit-model wrappers

Image and mask ops

File	What it adds
nodes_images.py	Image transforms (resize, pad, blend, …)
nodes_post_processing.py	Color, sharpen, blur, levels (40 KB)
nodes_mask.py	Mask ops
nodes_compositing.py	Layer compositing
nodes_image_compare.py	A/B compare nodes
nodes_canny.py	Canny edge detection (V3 example)
nodes_morphology.py	Morphology (dilate/erode/etc.)
nodes_color.py, nodes_curve.py	Color and curves
nodes_glsl.py	GLSL shader runner (37 KB)

Video, audio, 3D

File	What it adds
nodes_video.py, nodes_frame_interpolation.py	Video I/O + interpolation
nodes_audio.py, nodes_audio_encoder.py	Audio loading, encoding, encoders
nodes_load_3d.py	Load 3D meshes
nodes_webcam.py	Webcam capture

Training, dataset, primitives

File	What it adds
nodes_train.py	In-app fine-tuning workflows (53 KB)
nodes_dataset.py	Dataset construction nodes (53 KB)
nodes_primitive.py	Primitive UI widgets (Int, Float, String, …)
nodes_logic.py	Branching / boolean logic
nodes_math.py, nodes_number_convert.py, nodes_resolution.py	Math + helpers
nodes_string.py	String ops
nodes_replacements.py	Node replacement system
nodes_toolkit.py	Misc toolkit
nodes_painter.py	Painter widget
nodes_preview_any.py	Preview-any-type node
nodes_nop.py, nodes_rebatch.py	No-op, rebatching

Vision adapters

File	What it adds
nodes_sam3.py	SAM 3 segmentation (26 KB)
nodes_sdpose.py	SDPose human pose (35 KB)
nodes_rtdetr.py	RT-DETR detection
nodes_upscale_model.py, nodes_sdupscale.py	Upscaling
nodes_tomesd.py	Token-merge for SD
nodes_latent.py	Latent ops

Vendored dependencies

• comfy_extras/chainner_models/ — ESRGAN and friends, ported from chaiNNer.
• comfy_extras/frame_interpolation_models/ — RIFE and similar.

These are kept in-tree so users don't need separate pip installs.

Integration points

• Loaded by nodes.init_extra_nodes in nodes.py at startup.
• Imports comfy/ for actual ML work.
• Newer modules import from comfy_api/latest (V3 schema); older ones use V1 INPUT_TYPES.

Where to start a change

• Adding a new sampling tweak: model after nodes_pag.py or nodes_apg.py (small, focused).
• Adding a new model family's nodes: model after the most recent file (e.g., nodes_zimage.py, nodes_chroma_radiance.py).
• A complete worked V3 example is nodes_canny.py — only 40 lines including extension boilerplate.