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 > SimpleSyrup

ComfyUI Extension: SimpleSyrup

Repo Name

SimpleSyrup

Author
artificial-sweetener (Account age: 697 days)
Nodes
View all nodes(1)
Latest Updated
2026-10-11
Github Stars
0.0K

How to Install SimpleSyrup

Install this extension via the ComfyUI Manager by searching for SimpleSyrup
  • 1. Click the Manager button in the main menu
  • 2. Select Custom Nodes Manager button
  • 3. Enter SimpleSyrup 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

SimpleSyrup Description

SimpleSyrup adds ComfyUI nodes for model loading, tiled and contextual image sampling, region-specific prompts and detailing, object detection, tagging, and image and mask preparation.

SimpleSyrup Introduction

SimpleSyrup is a collection of custom nodes for ComfyUI. It brings several familiar image-making workflows into one pack: loading models, editing large images, prompting different parts of an image, detecting and refining objects, tagging image regions, and preparing images and masks.

If you have used A1111/WebUI, some ideas may feel familiar. For example, you can keep CLIP skip near your checkpoint loader, write separate prompts for detected regions using [SEP], and use tiled sampling when a large canvas is difficult to process. You do not need to use every feature together. SimpleSyrup nodes can be added to ordinary ComfyUI workflows as needed.

The author also uses SimpleSyrup in the first-party Base-Cubes pack for SugarSubstitute, a desktop front end for ComfyUI. Neither is required to use SimpleSyrup in ComfyUI.

How SimpleSyrup Works

A ComfyUI workflow is a graph: each node performs one job, and its outputs connect to the next job. SimpleSyrup adds nodes for common stages of that graph.

  • Load the parts your model needs. Some models need more than a single checkpoint. A SimpleSyrup loader can bring together the image-generation model, text encoder, VAE, and relevant settings. Think of it as gathering the right tools before starting a painting.
  • Describe the whole image and its parts. A global prompt describes the overall scene. Additional prompts can be paired, in order, with masks or detected regions. A clothing region, for example, can receive its own material description while the global prompt still guides the entire image.
  • Work on large canvases in smaller views. Tiled sampling examines overlapping sections of a large image, then combines their predictions. This can make a large canvas more manageable, but each section has limited awareness of the whole scene.
  • Keep the overall composition in view when needed. Contextual Diffusion combines detailed local views with a smaller view of the complete image during early sampling steps. In practical terms, the local views work on detail while the whole-image view helps them agree on the scene. It is intended for refinements such as clothing, color, materials, expression, and lighting while retaining the source composition.
  • Find and refine specific areas. Detection and segmentation nodes identify objects or turn masks into regions. Detailers can then work on those regions and blend the results back into the image.

The pack uses SEGS, a region format compatible with ComfyUI Impact Pack. A SEG carries information about a selected part of an image so compatible nodes can pass regions between detection, preview, prompting, tagging, and detailing steps.

SimpleSyrup Features

Model loaders

Simple Load Checkpoint loads a conventional checkpoint, offers an optional VAE override, and places CLIP skip alongside the other model controls.

The Anima, FLUX.1, FLUX.2, and Krea 2 loaders handle models that require specific supporting components. They offer manual component choices as well as automatic choices for known support files. Where supported, automatic choices can locate or download checksum-pinned files. You still need to select the appropriate diffusion model.

Some loaders let you choose weight precision and where text encoding runs. Moving text encoding to the CPU can reduce VRAM use, but it takes longer.

Large-image sampling

KSampler (Tiled Diffusion) divides a latent image into overlapping sections, evaluates them, and combines the results at each denoising step. MultiDiffusion averages overlapping predictions; Mixture of Diffusers gives more weight to the center of each section. Try this when the image benefits from smaller local evaluations and the overall scene remains consistent.

KSampler (Contextual Diffusion) adds an early whole-image correction to the local evaluations. It is useful when ordinary tiles preserve detail but disagree about the composition, such as repeating a figure or giving different sections of a building incompatible shapes. You can connect SEGS to guide where its local windows fall, and contexts_segs shows the windows it actually used.

Contextual Diffusion requires a tiled prediction pass and an additional smaller whole-image evaluation for each step that uses the correction. It currently rejects UniPC, regional conditioning, ControlNet, and GLIGEN because their behavior across both view sizes has not been validated. For a new pose, camera angle, or environment, the author recommends establishing the change at a normal working resolution before high-resolution refinement.

Tiled sampling is also available in Detail SEGS by Scale Factor w/ Tiled Diffusion and KSampler (Prompt by Tiled Region).

Regions, prompts, and detailers

  • Prompt SEGS w/ SAM finds objects using text with GroundingDINO, segments them with SAM, and can refine edges with ViTMatte. It also supports negative prompting.
  • Detect SEGS w/ Ultralytics creates regions from bounding-box or segmentation models. Confidence, label, and size filters help narrow what is selected.
  • Mask to SEGS turns an existing mask into regions. SEGS from SAM Output creates regions from automatic, unprompted SAM segmentation.
  • Simple Preview SEGS overlays regions on the image and lets you select them in an interactive grid. Batch SEGS combines ordered region inputs.
  • Scale-factor detailers enlarge one region crop, sample it, reduce it to its original size, and blend it back with feathering and optional denoise masks. Detail SEGS as Regions instead processes the full image in a MultiDiffusion pass with separate conditioning for each ordered region.

Encode Prompt Batch supports ADetailer-style [SEP] separators. Put the global prompt first, followed by one prompt for each corresponding mask or SEG. [SEP|name] can label a separator to make a long prompt easier to read; the matching still depends on order. For example, a global portrait prompt could be followed by a prompt for the jacket region and then one for the hair region. Keep the masks or SEGS in that same order.

Conditioning Batch Start, Conditioning Batch Append, and Batch Region Conditioning assemble ordered prompts from existing conditioning. Compose Regional Conditioning turns a global-first batch and ordered masks into standard masked ComfyUI conditioning. KSampler (Prompt by Region) and its tiled counterpart apply regional prompt batches while sampling.

If ComfyUI Prompt Control is available, Encode Prompt Batch w/ Prompt Control and Schedule & Encode Prompts add scheduled prompts and LoRA hooks across [SEP] regions. These optional nodes are not exported when Prompt Control is absent; the rest of SimpleSyrup still loads.

Tagging images and regions

Load WD14 Tagger loads a SmilingWolf WD14 ONNX tagging model and its tag list. Tag SEGS w/ WD14 tags existing region crops while keeping the resulting conditioning in region order. Tile & Tag SEGS creates an ordered set of tile regions, tags their crops, and returns regions paired with their conditioning.

For a configured OpenAI-compatible provider, Tag SEGS w/ External LLM sends each crop to a vision-capable model and returns aligned conditioning. External LLM Prompt sends system and user prompts and returns text; an image can also be supplied when the selected provider model supports vision.

Image, mask, latent, and sampling utilities

Load Image List preserves file selection order and keeps images as separate items, even when their dimensions differ. Load Mask Batch combines same-sized files into one mask batch and applies the chosen channel consistently.

Resize Image to Target offers stretch, keep-aspect, crop, and pad modes. You can choose crop or pad placement, round dimensions for divisibility, process batches in chunks, resize paired masks, and use GPU Lanczos resizing through TorchLanc. A crop fills the target area by removing excess image; a pad keeps the image content and fills the remaining space.

VAE Encode (Options) and VAE Decode (Options) expose ordinary and tiled VAE processing through an explicit tiling choice. Simple VAE Encode and Upscale Latent From Image can reuse an original latent when the workflow shows that the image came directly from an unchanged VAE decode. Editing, cropping, detailing, resizing, or loading an image breaks that connection, so these nodes cannot assume its original latent is still valid.

KSampler (Extras) provides additional sampler methods and schedules, including A1111-style Euler ancestral behavior, AYS SD1 and SDXL schedules, GITS, automatic_a1111, and RES4LYF options. These methods are also available in the contextual, tiled, and Attention Coupling KSamplers.

Seed Variation blends the sampler’s initial noise toward a second, repeatable seed. At strength 0, the initial noise stays with the sampler seed; at strength 1, it uses the variation seed’s initial noise. Ancestral and SDE samplers still use the sampler seed for additional noise introduced later.

Latent Diagnostics reports a latent’s shape, data type, device, and tiled-sampling compatibility without changing it. Scale Factor and Seed provide smaller workflow utilities.

SimpleSyrup Models

SimpleSyrup is a node pack, not a single image-generation model. Its loaders support several model families and supporting models. Choose a loader that matches the model and workflow you intend to use:

Model or componentWhat SimpleSyrup uses it for
Conventional checkpointsGeneral checkpoint workflows through Simple Load Checkpoint, with optional VAE override and CLIP skip.
AnimaAnima workflows using a Qwen text encoder and Qwen image VAE.
FLUX.1FLUX.1 workflows using CLIP-L, T5-XXL, and a VAE.
FLUX.2Workflows using FLUX.2 dev, Klein 4B, or Klein 9B/KV conditioning; the loader selects the matching text-encoder family for the chosen diffusion model.
Krea 2Raw or Turbo diffusion models using a Qwen3-VL 4B encoder with Krea layered conditioning and the Qwen Image VAE.
SAM and GroundingDINOFinding and segmenting objects into regions for prompts or detailers.
Ultralytics modelsCreating regions from object-detection boxes or segmentation results.
ViTMatteOptional edge refinement for segmented regions.
SmilingWolf WD14 ONNX modelsProducing tags from image crops or regions.
External vision-capable LLMsDescribing region crops or processing an optional image through a configured provider.

The Anima, FLUX, and Krea 2 loaders can resolve known supporting files automatically, but this does not mean they supply every diffusion model. For FLUX and Krea 2, you still install and select the diffusion model you want to use. The specific model and settings you choose affect the image; SimpleSyrup’s loader ensures the selected parts are assembled for that workflow.

Troubleshooting SimpleSyrup

A node is missing from the menu

  1. Check that you are using a current ComfyUI version; SimpleSyrup uses ComfyUI’s v3 extension API.
  2. Restart ComfyUI after adding or updating the pack.
  3. If the missing node is Encode Prompt Batch w/ Prompt Control or Schedule & Encode Prompts, check whether ComfyUI Prompt Control is available. Those nodes are optional.

A model is missing from a loader dropdown

  1. For SAM, GroundingDINO, ViTMatte, or WD14, check SimpleSyrup: Show downloadable models in loader dropdowns. When it is off, the dropdown shows models SimpleSyrup can verify locally.
  2. If you want a known downloadable entry, enable that setting and select the entry in the appropriate loader.
  3. For Anima, FLUX, or Krea 2 support components, try the loader’s automatic choice or select the component manually. These components are handled by their own loaders rather than that dropdown setting.
  4. Confirm that you have separately selected the diffusion model required by your workflow.

Regional prompts affect the wrong area

  1. Check that your first prompt is the global prompt.
  2. Compare each later [SEP] prompt with the order of the masks or SEGS connected to the workflow.
  3. Use Simple Preview SEGS to see which region each ordered item represents.
  4. Reorder the prompts or regions so each description matches its intended area. Names in [SEP|name] help you read the batch, but do not change order-based matching.

Tiled output has repeated or inconsistent objects

  1. Inspect the complete image, not just individual tiles or close-up details.
  2. If ordinary tiled sampling keeps detail but loses the scene’s overall consistency, try KSampler (Contextual Diffusion) for an edit that should preserve the existing composition.
  3. Use contexts_segs to inspect the local windows evaluated by Contextual Diffusion.
  4. If your goal is an entirely new pose, viewpoint, or environment, establish that change at a normal working resolution before refining it at high resolution.

A latent cannot be reused

Simple VAE Encode and Upscale Latent From Image can trace an original latent only when the image comes directly from an unmodified VAEDecode in the workflow. If you loaded, edited, cropped, detailed, or resized the image afterward, use a normal encode path instead.

External LLM image tagging does not work

  1. Check the SimpleSyrup: External LLM endpoint and SimpleSyrup: External LLM API key settings.
  2. Save the settings so the available provider models refresh in connected nodes.
  3. For image inputs, select a provider model that supports vision. Text-only models cannot process a region crop or optional image.

Frequently asked questions

Do I need SugarSubstitute or Base-Cubes?
No. SimpleSyrup works as a node pack in ordinary ComfyUI workflows.

Do I need ComfyUI Impact Pack to understand SEGS?
SEGS is the shared region format used for compatible workflows. SimpleSyrup provides its own region-related nodes and can pass compatible SEGS between its nodes and Impact workflows.

Should I always use Contextual Diffusion for a large image?
No. Start with tiled diffusion when overlapping local views are sufficient. Contextual Diffusion adds a whole-image evaluation during its correction steps and is aimed at edits where local views lose track of the composition.

Will an automatic loader download the diffusion model for me?
Do not assume so. The documented automatic choices concern known supporting text encoders and VAEs; you still select the diffusion model.

Learn More about SimpleSyrup

SimpleSyrup is licensed under AGPL-3.0-or-later. For the origins and terms of included third-party components, see the third-party notices and RES4LYF license copy.

SimpleSyrup 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.

SimpleSyrup detailed guide | ComfyUI