Skip to content

About

Utility nodes for ComfyUI: logic, mask, image, PostFx film looks, dataset analysis (Caption Audit, Image Quality Gate), Social Media Export, SeedVR2 streaming, loaders and workflow helpers (AutoBypass, Anything Everywhere, comparers).

Resources

Stars

2 stars

Watchers

0 watching

Forks

Repository files navigation

ComfyUI-BCNodes

Utility nodes for ComfyUI, in one small pack.

Registration key Display name What it does
BC_AutoBypass Auto Bypass Bypasses its targets automatically when a watched source is empty
BC_LogicBoolean Logic Boolean 0–1 float → BOOLEAN / NUMBER / INT / FLOAT
BC_IsMaskEmpty Is Mask Empty MASK → BOOLEAN
BC_MaskFillHoles Mask Fill Holes Fills enclosed holes in a mask
BC_MaskGrow MaskGrow Grows / shrinks a mask, then blurs it
BC_DrawMaskOnImage Draw Mask On Image Paints a colour (with opacity) through a mask onto an image batch
BC_BlockifyMask Blockify Mask A mask as the blocks of its bounding box that hold any of it
BC_RepeatMaskBatch Repeat Mask Batch A mask batch repeated amount times
BC_ImageScaleByAspectRatio Image Scale By Aspect Ratio Scales an image / mask to an aspect ratio and side length; MASK output is never None
BC_ImageResize Image Resize Resizes an image batch (and mask) to a width and height: stretch, keep the proportion, pad or crop
BC_JoinImageLists Join Image Lists Concatenates image lists, unlimited inputs
BC_MathExpression Math Expression Arithmetic over a, b, c without eval()
BC_PromptList Prompt List One prompt per line, as a list
BC_AspectPromptList Aspect Prompt List Prompts grouped under [WxH] headers, as prompt / width / height lists
BC_AnySwitch Any Switch First connected non-None input, any type, unlimited inputs
BC_SelectSwitch Select Switch Input of the selected named option, any type; only the selected branch runs
BC_Seed Seed Seed widget; -1 draws a new random seed on every run
BC_ShowText Show Text Shows incoming text on the node, passes it on
BC_QwenLM Qwen LM Runs a Qwen chat model (Qwen3.5-4B / 9B, Qwen3.8-27B, Qwen3-VL-8B, and abliterated builds of Qwen3.5-9B and Qwen3-VL-8B) through ComfyUI core: system / user / prefill and images in, text and thinking out, with the official chat templates and sampling defaults and one LoRA; the model file is found in models/Qwen-LM or models/text_encoders, or downloaded on first use
BC_LMConfig LM Config Sampling settings for an LM node: only the fields you change or link replace the model's official defaults
BC_LMLoader LM Loader Picks the chat model a wrapper node runs (Captioner, Prompter, PreFlight Observe): family, model, precision, one LoRA and its strength, keep loaded; loads nothing itself, and its lists follow the selected family and model
BC_Captioner Captioner One training caption per image frame with the model of a connected LM Loader: a built-in tags or prose prompt, or the image alone (none) for your own image-to-caption LoRA, optionally with an instruction or an LM Config; outputs the frames and their captions as lists, ready for Save Image With Caption
BC_Prompter Prompter One text-to-image prompt from a short brief and / or a reference image with the model of a connected LM Loader: a built-in prompt that writes the person's appearance (include identity) or leaves it to a character LoRA (exclude identity), or none for your own prompter LoRA; text out, and with a CLIP connected the positive conditioning as CLIP Text Encode encodes it
BC_PreFlightObserve PreFlight Observe Runs the LM Loader's chat model (default Qwen3.5-9B INT8 ConvRot) as a pure observation sensor: an image or frame batch → a strict observations JSON (garment, exposure, framing, pose, in-image text, …); greedy, so the same input gives the same output; an error comes back as {"error": ...}
BC_PreFlightReport PreFlight Report The rules engine: observations + caption → per-platform verdict ranges (Instagram / TikTok / X: OK, RISK = reach demotion, BLOCK = removal), a summary and the prediction logged to the feedback store
BC_PreFlightOutcome PreFlight Outcome Records what a platform actually did to a published post (clean / demoted / removed); its record list shows each new prediction at once
BC_PreFlightCalibrate PreFlight Calibrate The feedback store as two calibration tables (per platform, per rule), as text for Show Text; runs again only when the store changed
BC_ImageComparer Image Comparer Two images on the node, compared with a sliding divider
BC_PowerLoraLoader Power Lora Loader MODEL + any number of LoRA rows → MODEL, no CLIP; each LoRA with the key fix of Lora Loader (Key Fix)
BC_LoraLoaderKeyFix Lora Loader (Key Fix) MODEL + one LoRA → MODEL, with the keys ComfyUI's own loader leaves out renamed first (lightx2v .diff_m, PEFT keys without the diffusion_model. prefix); the console says how many were renamed and how many match no module
BC_PatchSageAttention Patch Sage Attention MODEL → a clone whose attention runs one SageAttention kernel, for that model only (no global --use-sage-attention); auto takes the most accurate kernel, qk_quant its Q·K quantization (per_thread / per_warp), and a call the kernel cannot take, a kernel error or a NaN output runs the original attention; the console counts both per run
BC_AnythingEverywhere Anything Everywhere Feeds unconnected inputs of a type at prompt time
BC_FastGroupsBypasser Fast Groups Bypasser One bypass toggle per group
BC_BiRefNetRemoveBackground BiRefNet Remove Background Background removal with BiRefNet, plain torch
BC_DepthAnythingV2 Depth Anything Depth map with Depth Anything V2 Small or Depth Anything 3 (Small, Base, Mono-Large, Metric-Large; all Apache-2.0), near = white, at a ControlNet preprocessor's size or exactly width x height
BC_SeedVR2FramingDownscale SeedVR2 Framing Downscale SeedVR2 Resize's downscale_factor from the face size: the tallest SAM 3 face box picks close-up, medium or far, each with its factor; the measured face fraction comes out too
BC_SeedVR2Resize SeedVR2 Resize Original image → the padded frame SeedVR2 encodes (lanczos downscale, shortest-edge antialiased bicubic, pad 16, 4n+1 frames) plus the colour reference
BC_SeedVR2VAEEncode SeedVR2 VAE Encode SeedVR2 VAE encode with the frames streamed from RAM slice by slice, so VRAM does not grow with the frame count
BC_SeedVR2ChunkSize SeedVR2 Chunk Size The longest 4n+1 chunk the SeedVR2 sampler fits on this card, from the latent's size and the card's total memory; wire to Split SeedVR2 Latent's manual frames_per_chunk
BC_SeedVR2VAEDecode SeedVR2 VAE Decode SeedVR2 VAE decode with every decoded slice streamed to RAM, so VRAM does not grow with the frame count
BC_SeedVR2PostProcess SeedVR2 PostProcess Post-Process SeedVR2 Output one frame at a time into one float16 output, no full-size temporaries
BC_SeedVR2PreprocessCompact SeedVR2 Preprocess (Compact) SeedVR2 Resize and VAE Encode in one node: outputs only the latent and the plan PostProcess (Compact) reads, no colour reference kept
BC_SeedVR2PostProcessCompact SeedVR2 PostProcess (Compact) SeedVR2 VAE Decode and PostProcess in one node: decodes only the kept frames and corrects them in place against a reference rebuilt per chunk from the original frames
BC_AutoModelDownloader Auto Model Downloader Lists a workflow's models and fetches the missing ones into models/
BC_PostFxApply PostFx Apply Applies a postfx film-emulation look (theme + condition + strength) to an image batch, optional look override and mask
BC_PostFxTheme PostFx Theme Built-in postfx theme → POSTFX_LOOK, to start a chain from a named look
BC_PostFxCustomLook PostFx Custom Look Builds a look from common controls, or overrides them on top of an incoming look
BC_PostFxLut PostFx LUT 3D .cube LUT as a look, standalone or layered on a theme
BC_PostFxSignatureSheet PostFx Signature Sheet Labeled contact sheet of every theme in a category applied to one image
BC_CaptionAudit Caption Audit Runs caption-audit over a LoRA caption folder and draws the report card on the node
BC_SocialMediaExport Social Media Export One platform-ready derivative per ticked platform, minimum crop, spec-driven
BC_ImageQualityGate Image Quality Gate Blur / sharpness / noise / clipping / entropy → PASS / SO-SO / FAIL badge, verdict and scores
BC_SaveImage Save Image Saves images with folder / file names built from prompt widget values, any Pillow format, prompt + workflow embedded; preview only in the gallery, never under the node
BC_SaveImageWithCaption Save Image With Caption Saves images as PNG and, with a caption connected, the caption next to each image under the same name, for training datasets
BC_SkinTexture Skin Texture Micro-texture on skin inside a SAM 3 mask: boosts the image's own detail and multiplies in a synthetic pore field, in linear light
BC_FrequencyMerge Frequency Merge The low frequencies (structure, colour) of one image + the high frequencies (texture, fine detail) of another of the same size, split by a Gaussian
— Align Align / distribute buttons in the selection toolbox
— Process Monitor Not a node: live RAM / VRAM bars, an estimate before a run, per-node measurement, the reason a killed run died, a full clear back to the startup memory without a restart, and a drop stale outputs setting (off) (see Process Monitor)

Registration keys are BCNodes' own, so the packages above can be installed side by side without a clash. Type BCNodes in the node library to see them all; in the menu they sit in these groups:

Category Nodes
BCNodes/logic Logic Boolean, Math Expression, Any Switch, Select Switch, Seed
BCNodes/mask Mask Fill Holes, MaskGrow, Draw Mask On Image, Blockify Mask, Repeat Mask Batch, Is Mask Empty, BiRefNet Remove Background
BCNodes/image Image Scale By Aspect Ratio, Image Resize, Join Image Lists, Depth Anything, Social Media Export, Save Image, Save Image With Caption, Skin Texture, Frequency Merge
BCNodes/postfx PostFx Apply, Theme, Custom Look, LUT, Signature Sheet
BCNodes/analysis Image Quality Gate, Caption Audit
BCNodes/text Prompt List, Show Text
BCNodes/lm Qwen LM, LM Config, LM Loader, Captioner, Prompter
BCNodes/preflight PreFlight Observe, Report, Outcome, Calibrate
BCNodes/loaders Power Lora Loader, Lora Loader (Key Fix), Auto Model Downloader
BCNodes/model Patch Sage Attention
BCNodes/seedvr2 SeedVR2 Framing Downscale, Resize, VAE Encode, Chunk Size, VAE Decode, PostProcess; Compact: Preprocess (Compact), PostProcess (Compact)
BCNodes/workflow Image Comparer, Anything Everywhere, Fast Groups Bypasser, Auto Bypass

Installation

cd ComfyUI/custom_nodes
git clone https://github.com/0xBeycan/ComfyUI-BCNodes
pip install -r ComfyUI-BCNodes/requirements.txt

Restart ComfyUI. requirements.txt holds opencv-python (the morphology of MaskGrow and the hole filling of Mask Fill Holes) and the two packages that back the PostFx and Caption Audit nodes, postfx and caption-audit; everything else ships with ComfyUI. Without postfx and caption-audit the pack still loads and only those nodes are absent. Patch Sage Attention needs SageAttention 2.2.x, which is not in requirements.txt (see Model); without it the node leaves the model on ComfyUI's attention. If this repository is still present under its old name ComfyUI-AutoBypass, delete that folder — its AutoBypass node is this pack's BC_AutoBypass.

The Align buttons are not a node; they appear in the toolbox above a multi-selection.

Documentation

  • Logic — booleans, arithmetic, switches and the seed
  • Mask — mask checks, fill / grow / draw / blockify / repeat, BiRefNet background removal
  • Image — scaling and resizing, image lists, depth maps, social media export, saving, skin texture, frequency merge
  • PostFx — film-emulation looks: themes, custom looks, LUTs, signature sheets
  • Analysis — the caption set audit and the image quality gate
  • Text — prompt lists and showing text
  • LM — chat models through ComfyUI core: Qwen LM, LM Config, LM Loader, Captioner, Prompter, the models and their files, sampling, LoRAs, models.yaml
  • PreFlight — how Instagram, TikTok and X would likely treat an image or video before it is published: the observation sensor, the rules engine, reading a verdict, the feedback loop and calibration
  • Loaders — LoRA loading (with the key fix for LoRAs core loads only in part) and the model downloader
  • Model — SageAttention for one model in one workflow: the modes, qk_quant, long sequences, the calls that run the original attention, nan_guard, the console, install, an A/B test
  • SeedVR2 — the downscale factor from the framing, the SeedVR2 input stage, streaming VAE encode / decode, the chunk size and the VAE tile for the card, post-processing
  • Workflow — image comparer, Anything Everywhere, group bypass, Auto Bypass, the Align buttons
  • Unused outputs — unconnected whole-batch outputs come out empty
  • Measurements — time, RAM and output size of the mask nodes and Image Resize against the nodes they replace
  • Process Monitor — RAM / VRAM live, an estimate before a run, per-node measurement, the report of a killed run, the full clear
  • Development — the repository layout, the tests and the import rule

Third-party code

Everything else in this pack is BCNodes' own code.

License

MIT. The code in models/birefnet/arch/ keeps its own MIT notice, in models/birefnet/arch/LICENSE; the code in models/depth_anything_v2/arch/ keeps its Apache-2.0 license, in models/depth_anything_v2/arch/LICENSE.

About

Utility nodes for ComfyUI: logic, mask, image, PostFx film looks, dataset analysis (Caption Audit, Image Quality Gate), Social Media Export, SeedVR2 streaming, loaders and workflow helpers (AutoBypass, Anything Everywhere, comparers).

Resources

Stars

2 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages