Save 4 hours! We auto-setup your workflow! Free!

Drop your workflow.json — we handle every dependency, custom node, and model. Just open the link and run.

Auto-Setup Workflow Json (Free) Now!
ComfyUI > Nodes > toobusy

ComfyUI Extension: toobusy

Repo Name

toobusy

Author
nicekriss (Account age: 1318 days)
Nodes
View all nodes(3)
Latest Updated
2026-10-25
Github Stars
0.02K

How to Install toobusy

Install this extension via the ComfyUI Manager by searching for toobusy
  • 1. Click the Manager button in the main menu
  • 2. Select Custom Nodes Manager button
  • 3. Enter toobusy in the search bar
After installation, click the Restart button to restart ComfyUI. Then, manually refresh your browser to clear the cache and access the updated list of nodes.

Visit ComfyUI Online for ready-to-use ComfyUI environment

  • Free trial available
  • 16GB VRAM to 80GB VRAM GPU machines
  • 400+ preloaded models/nodes
  • Freedom to upload custom models/nodes
  • 200+ ready-to-run workflows
  • 100% private workspace with up to 200GB storage
  • Dedicated Support

Run ComfyUI Online

toobusy Description

toobusy is a ComfyUI custom-node collection that folds repetitive image and video workflows into compact nodes for generation, editing, retouching, upscaling, reference planning, and video enhancement.

toobusy Introduction

toobusy is a collection of custom nodes for ComfyUI. Its guiding idea is to fold the graph: replace a long chain of repetitive steps with a smaller number of purpose-built nodes. You can spend less time connecting loaders, samplers, image-processing steps, and outputs, and more time making creative decisions.

The collection covers several kinds of work:

  • Image creation and editing: Generate with Flux2 Klein, Z-Image Turbo, or a local Ideogram4 model; prepare references; refine details; and upscale.
  • Video creation and enhancement: Extend Wan SCAIL-2 clips, process longer Wan Animate 2 sequences, or enhance video with FlashVSR.
  • Planning and visual direction: Organize reference images, draft prompts, build layouts, plan keyframes, and work on a storyboard or rough painting inside ComfyUI.
  • Targeted retouching: Improve fine texture in skin, hair, and clothing, or revisit detected subjects in a large image without entering crop coordinates by hand.

A compact graph does not remove the need for the underlying models and supporting nodes. Each workflow still needs the components listed in its example or requirements. The source material describes different features and examples from different releases, so check the requirements for the particular workflow you open.

How toobusy Works

Think of a normal ComfyUI graph as a production line: one node loads a model, another prepares the prompt, another samples an image, and another decodes it. toobusy packages selected production lines into single nodes. You still choose the inputs and creative settings, but you do not have to rebuild every intermediate connection.

Some workflows instead connect a few specialized nodes:

  • Reference-driven images: Put images, text, audio, and LoRAs on a toobusy Reference Board. The board packages them into a TOOBUSY_BUNDLE, which a Prompt Director and compatible generation nodes can use.
  • Selective detail work: Detect a face or person, crop each detected area, process those crops at a useful working size, then blend them back into the original image.
  • Long video: Divide a sequence into manageable chunks, process the chunks, and join them into the requested result.

For example, a character-sheet workflow can establish a three-view character first, then use that result as the reference for a second pass with poses or props. This separates the decision about who the character is from the decision about what the character does.

Nodes are grouped in ComfyUI under toobusy/Plan, toobusy/Make, toobusy/Image, and toobusy/DreamID.

toobusy Features

Reference boards and prompt direction

toobusy Reference Board collects image, audio, text, and LoRA cards in one place. You can name and replace card images, preview them on hover, drag media out, and save or load board presets. Image-card modules can remove a background, erase a face, or keep only a face; these modules have optional dependencies.

toobusy Flux2 Klein Prompt Director uses the connected board’s registered cards to assemble a final prompt through buttons. Its FaceSwap, Product Swap, and Character Swap choices help determine how references are used. Camera, lighting, and style chips give you additional direction. Bundle Unpack and Bundle Get let other parts of a graph retrieve specific items from the bundle.

Example: Keep a face reference and a clothing reference on the board, then change your prompt direction without repeatedly reconnecting those images. The toobusy Flux2 Klein node can take the bundle directly and follow the Director’s swap choice when its reference order is set to auto.

Image generation and layout

  • toobusy Flux2 Klein: Combines the Flux2 Klein 9B reference-generation chain. It supports up to five reference slots and five LoRA slots. The first reference normally informs output size; size_mode can instead enforce a ratio or manual size.
  • toobusy Z-Image Turbo: Combines text-to-image, image-to-image, and latent-input paths. Override inputs and passthrough outputs make it possible to connect further processing without rebuilding the full generation chain.
  • toobusy ZIT ControlNet: Adds depth, edge (canny), or pose guidance to a Z-Image Turbo workflow. Use these when the composition or body position matters as much as the prompt.
  • toobusy Hires Upscale: Combines an upscale-model pass, resizing, and VAE encoding for a higher-resolution refinement loop.
  • Ideogram workflow: toobusy Ideogram Prompt Polish turns a Korean scene description into a structured prompt. With an image and a compatible vision model connected, it can also draft layout JSON from one image. Ideogram Layout Builder lets you inspect or adjust the composition before toobusy Ideogram4 T2I generates an image with a local Ideogram4 model, not a web API. toobusy Layout Text Overlay can place editable, sharply rendered Korean text over a generated image when the image model does not render that text reliably.

The Ideogram scene-to-image workflow shows these steps together:

Prompt Polish, Layout Builder, and Ideogram4 workflow

Retouching, detection, and upscaling

2BZ SeedVR2 Tile Restore restores fine texture in tiles and blends the tiles back into an image at its input size. The default tile_megapixels is 2; reducing it lowers the amount of image processed at once and can help on cards with less VRAM. The supplied guidance suggests starting around 1 for 8 GB, 1.5–2 for 12 GB, 2–3 for 16 GB, and 3–4 for 24 GB. Images smaller than the default upscale_small_to value of 3 MP are enlarged for restoration and then returned to their original size.

2BZ Retouch Blend (skin-aware) mixes the restored fine detail with the original image’s broader shapes and colors. It can apply a gentler blend inside a detected skin mask. This is useful when you want more believable hair or fabric without pushing facial texture as strongly. The SeedVR2 any-size retouch workflow includes stronger and gentler output branches.

Two related workflows serve different goals:

  • Automatic face-size retouch: 2BZ Face-Size Scale Plan measures detected faces and plans a working size around a 200-pixel face. 2BZ Use Upscaled If Needed skips VOSR 2.0 when enlargement is unnecessary. The final image stays at the enlarged working size. See the automatic-size retouch workflow.
  • VOSR 2.0 retouch: Uses VOSR 2.0 as a gentler restoration engine with 2BZ Retouch Blend. The source recommends SeedVR2 as the faster, more pronounced default, and VOSR when preserving the original look is more important. The VOSR retouch workflow returns a retouched image; the separate VOSR upscale workflow saves an image enlarged by its chosen upscale factor.

For a large scene with several people, 2BZ Crop Each BBox (list) can crop detected subjects, resize each crop toward target_megapixels (default 4.0), and send them through a separate detail pass. 2BZ Paste Crops Back returns them to their original positions, blending their edges with feather_px (default 24). A detection prompt such as person:8 requests up to eight people; without :N, the described SAM3 prompt format detects one per category. See the Qwen-Image 2.1 auto-detail workflow.

Video workflows

  • toobusy Wan SCAIL Extend Sampler: Combines Wan 2.1 SCAIL-2 generation and extension. In target total frame mode, you specify the desired total and let the node plan the base and extension chunks; manual segments retains direct segment control. color_anchor, color_match_strength, and color_sample help manage color changes between chunks. See the workflow and result preview.
  • toobusy Wan Animate 2 Long Sampler: Splits a driving video into valid 4k+1-frame chunks, joins successive chunks with a shared frame, and trims the result to total_frames. Suggested starting values for frames_per_sampler are 33 frames at 12 GB, 49 at 16 GB, and 81 at 24 GB; these are starting points, not guarantees for every setup.
  • FlashVSR v1.1 Full + BSA: Provides a VRAM and resolution preset, loader, long sampler, and full decoder for chunked video enhancement. Presets choose different offloading and decoding approaches: lower-VRAM options prioritize fitting the job, while faster options may use more VRAM. Decoder tile_preset choices are safe, balanced, and fast. Models are not downloaded automatically. See the FlashVSR long-video workflow.

Planning and utility nodes

toobusy Keyframe Maker converts a brief and references into shot beats and keyframe prompts; its results depend on the connected text-generation model. toobusy Storyboard Board supports image boards and keyframe marking inside ComfyUI, while toobusy Paint Canvas lets you make a rough painting or mask before generation. toobusy Load CLIP loads supported safetensors or .gguf text encoders; .gguf requires ComfyUI-GGUF. Background removal and face masking are available as optional image-processing nodes. DreamID-Omni Loader/Talker nodes delegate talking-head work to an installed DreamID-Omni extension.

toobusy Models

toobusy is not one image model. It is a node collection that connects to different models depending on the workflow. Model files are not included in the repository; a workflow may also require ComfyUI support nodes.

WorkflowModel or componentWhen to use it
Flux2 KleinFlux2 Klein 9B, a compatible Flux2 text encoder, and Flux2 VAECreate images guided by reference images, including character or product variations.
Z-Image TurboZ-Image Turbo diffusion model, lumina2 text encoder, and VAEGenerate from text, transform an image, or continue from a latent.
MiniMax H3 single imageMiniMax H3 Ref2V and MiniMax H3 Image VAEEdit a single still image using H3’s video-and-audio latent format.
SeedVR2 retouchSeedVR2 3B int8 restoration model and SeedVR2 VAEAdd more noticeable fine texture while keeping the original image size.
Skin or face detectionSAM3.1 multiplex checkpointIdentify regions for skin-aware blending or detection-led detail work.
VOSR 2.0VOSR 2.0 and a compatible VOSR nodeChoose subtler retouching or a separate enlargement workflow.
Local Ideogram4Ideogram4 models and ComfyUI Ideogram4 support nodesBuild a layout and generate an image locally.
FlashVSRFlashVSR v1.1 model and its other required assetsEnhance longer videos in chunks.

Other named workflows, including Wan SCAIL-2, Wan Animate 2, LTX2.3, and DreamID-Omni, likewise need their corresponding models and support nodes. Open the relevant example workflow to see its complete set of components.

MiniMax H3 note: toobusy MiniMax H3 Image Latent makes a single-frame (T=1, batch-one) H3-compatible latent. A regular Empty Latent Image is not a substitute. For video rather than a still image, use the official Empty MiniMax H3 AV Latent. See the six-edit single-image example.

What's New with toobusy

The supplied readme identifies itself as v0.4.11, but it also describes workflows with later minimum-version requirements. The entries below reflect only changes explicitly identified in the supplied material; they are not a claim that v0.4.11 is the latest release.

  • v0.2.10: Added Flux2 Klein and Load CLIP; expanded Z-Image Turbo’s override and passthrough connections; added Hires Upscale and ZIT ControlNet; and added SCAIL color_anchor. These changes made reference generation and follow-on refinement easier to keep in a compact graph.
  • v0.2.11: Added target total frame planning and more SCAIL color-matching controls, and refreshed the SCAIL-2 example. You can request a final frame count without planning every extension yourself.
  • v0.3.0 / v0.3.1: Introduced the Reference Board → Prompt Director → Flux2 Klein workflow, bundle extraction nodes, optional background and face tools, DreamID-Omni integration, and bundle-aware Flux2 Klein reference ordering.
  • v0.5.2: Moved the ABC score studio to the separate toobusy-abc-studio project. It is a music-score tool, rather than a current feature of the main toobusy node collection.
  • v0.5.5 and later: The source specifies v0.5.5 or newer for its VOSR 2.0 retouch workflow; it warns that v0.5.4 can introduce stair-step artifacts when reducing a large restored result.
  • v0.5.6 and later: The source specifies v0.5.6 or newer for the automatic face-size retouch workflow.

No separate version-by-version history is supplied for the other workflows described here.

Troubleshooting toobusy

A node in an example workflow is missing. What should you check?

  1. Check whether the missing node belongs to toobusy or to another extension named in the workflow.
  2. Check your toobusy version. The supplied readme warns that, at the time it was written, ComfyUI Manager could install registry version 0.4.8 because versions 0.4.9 and 0.4.10 were unavailable there. In that situation, MiniMax H3 character-sheet nodes added in 0.4.9 would not appear.
  3. After updating, restart ComfyUI. If controls or buttons still look old, hard-refresh the browser page.

A workflow opens but fails when it runs. Why?
A folded node still calls underlying models or ComfyUI functionality. Read the workflow’s Note nodes and check that its required model files, support nodes, and optional dependencies are present. For example, local Ideogram4 requires Ideogram4 support in ComfyUI; it does not switch to an online API when that support is missing.

MiniMax H3 rejects my latent. What should I use?
For the single-image workflow, connect toobusy MiniMax H3 Image Latent instead of a normal Empty Latent Image. For video, use the official H3 AV latent node. The single-image example shows the intended connections.

Retouching runs out of VRAM. What can I change?
In the SeedVR2 tile workflow, reduce tile_megapixels so each tile is smaller. In a VOSR workflow, review its working size and vae_tile_size. In FlashVSR, choose the VRAM preset and decoder tile preset appropriate to your available memory. Smaller or more cautious settings may take longer.

My retouch result looks too strong or too subtle. Where should I adjust it?
Compare the stronger and gentler branches in the SeedVR2 retouch example, including their skin-aware blend settings. If you want a lighter effect that stays closer to the original, review the VOSR 2.0 retouch example. A retouch workflow is different from the separate upscale workflow, which changes output dimensions.

Why did a detected crop or reference not appear?
Check the detector prompt and its limit, such as face:4 or person:8. For MiniMax H3 character sheets, an optional reference can be disabled, in which case its image branch is skipped. Semantic Reference can retain an image’s descriptive analysis while passing only images judged safe for direct reference; Reference Manifest updates the <Picture N> labels to match the images actually passed. See the two-stage character-sheet example.

Can I use ABC score studio through toobusy?
It moved to toobusy-abc-studio beginning with v0.5.2. If you use an older toobusy version that still contains it, running both copies can attach duplicate buttons. The separate project provides its own documentation and audio examples.

Learn More about toobusy

The most practical starting points are the supplied workflows: open one that matches your goal, read its Note nodes, and follow its connections before changing settings.

toobusy Related Nodes

RunComfy
Copyright 2025 RunComfy. All Rights Reserved.

RunComfy is the premier ComfyUI platform, offering ComfyUI online environment and services, along with ComfyUI workflows featuring stunning visuals. RunComfy also provides AI Models, enabling artists to harness the latest AI tools to create incredible art.

toobusy detailed guide | ComfyUI