WorkflowX-Configurator Introduction
WorkflowX-Configurator, called WorkflowX in the ComfyUI interface, is a toolkit for building and reusing image, video, and audio workflows. It adds 42 active nodes, three sidebar libraries, and right-click canvas tools.
A ComfyUI workflow is a connected set of steps, or nodes. WorkflowX helps you change those steps without rebuilding the whole graph. You can switch between saved model choices, apply an ordered set of LoRAs, reuse prompts and groups of nodes, select a workflow configuration, and route the selected settings to the nodes that need them. It also provides tools for loading and editing media, preparing prompts, using supported image-generation APIs, and saving video.
You do not need every feature for every project. For example, you might use the image loaders and comparison editor for a small editing workflow, then add Config SelectorX when you want one graph to support both a quick draft and a final render.
How WorkflowX-Configurator Works
Think of a workflow as a studio production line: one node supplies an image or setting, another changes it, and a later node previews or saves the result. WorkflowX adds tools for both the material moving through the line and the choices that control the line.
- Media nodes bring in images, video, or audio and pass usable outputs to later nodes.
- Model and LoRA nodes choose the diffusion model and apply enabled LoRAs in the order shown on the node.
- Prompt nodes turn descriptions, reference images, or structured fields into text for a generation workflow.
- Config SelectorX chooses a named setup and controls which native ComfyUI groups participate. Matching Set/Get nodes and relays supply values from the selected setup.
- Library and canvas tools help you store useful work and rearrange an existing graph.
For example, you could place draft and final settings in two named groups. A Get Dimensions node can then supply the width and height selected for the current configuration. Switching configurations changes the settings the workflow uses, without requiring you to reconnect its downstream nodes. The configuration and routing example demonstrates this pattern.
WorkflowX saves many interface choices in the workflow, but it does not bundle your source media or model files with the workflow. When you open an example or share a graph, you may need to select files available on that computer. Keep API credentials and machine-specific paths out of workflows you intend to share.
WorkflowX-Configurator Features
Configure and route a workflow
Config SelectorX manages named configurations using native ComfyUI groups. Its Scopes controls decide whether a group is controlled by a configuration, muted, bypassed, or ignored. Its Configs controls let you choose an Active, Bypass, Mute, or Ignore state for each controlled group. A useful starting point is a Draft configuration for experiments and a Final configuration for the settings you want to render with.
Typed Set/Get pairs carry keyed integers, decimal values, text, booleans, sampler choices, and scheduler choices. Other pairs route width and height together or pass a connected object, such as an image, to another part of the workflow. Set/Get Reference also offers mute controls for reference branches.
Use a distinct, matching key for each value you intend to route. A key is the label connecting a Set to its Get; for example, a Set Dimensions and Get Dimensions pair might both use output_size. If more than one eligible Set competes at the same priority level, resolution stops with an error rather than silently choosing one. Group names must also be unique across the reachable workflow. See the Config SelectorX guide for the full selection rules.
Load images, video, and audio
- Load ImageX selects an image from ComfyUI’s input directory and provides both the image and a mask. Its thumbnail browser makes it easier to find a file visually.
- Load ImageX Adv adds image resizing, cropping, mask controls, and width and height outputs. Its editor choices are saved as workflow state, but the source image file is not embedded. See the advanced image-loader guide.
- Load VideoX Adv offers a visual timeline for choosing a segment, plus crop and geometry controls. It can output video, frames, audio when present, and final dimensions. See the video-loader guide.
- Load AudioX lets you select, listen to, trim, convert, and process audio. Its recovery controls can help with certain damaged files without changing the source file. See the audio-loader guide.
- Voice ChangerX adjusts qualities of one spoken voice, including pitch, brightness, breathiness, and gain. An optional reference voice can suggest approximate slider settings; ordinary runs then use the settings you choose. Neutral settings return the original audio. See the voice guide.
As a practical example, use Load VideoX Adv to select a short portion of a clip before sending its frames into an image-processing branch. Use Load ImageX Adv when an edit needs a specific crop and its corresponding mask.
Choose models and manage LoRAs
Load Diffusion Model X keeps several preferred diffusion-model filenames visible in one node, with one active choice at a time. Its model picker can use LoRA Manager previews and metadata when available, but it also works with ComfyUI’s model folders alone. Changing the active row changes which model the node loads. See the model-loader guide.
LoraX applies enabled LoRAs in the order displayed. You can search, add, remove, reorder, or disable rows and set strengths for the model and optional CLIP text encoder independently. Its outputs include trigger words when available and a readable record of the applied stack. Load Lora PickerX offers a Windows file picker and can show related checkpoints from the same directory; paths outside the usual model folders remain specific to that computer. See the LoraX guide and Load Lora PickerX guide.
Unload Models By Type can request the release of selected loaded components after a chosen workflow stage. Its trigger connection establishes when that request runs. This is useful in a longer graph that no longer needs a model before a later output step.
Build and organize prompts
Unified Autoprompter X helps compose prompts from your instructions, optional images, text, or layout data. You can choose a target-model profile, a generation path, and natural-language, tag, or JSON output. Other controls include detail level, optional negative prompts, reusable @tag presets, and a choice of generating on demand or when the workflow is queued. It supports several listed remote and local prompting providers; provider credentials and tuning remain browser-local. Connected images can inform the prompt, but they do not themselves choose the downstream generation path. See the Autoprompter guide.
The JsonX tools provide a more structured approach:
- Visual Builder lets you fill prompt fields and apply presets.
- Prompt Template Importer analyzes a source prompt and reports which fields it accepted or skipped.
- Template Randomizer makes repeatable, seed-controlled choices from template rules and records what it changed.
Use these tools when you want prompt variations with a consistent structure, or when you need to inspect how a complex prompt was assembled. See the JsonX user guide and prompt-toolchain example.
Generate or edit through image APIs
WorkflowX includes NanoBanana Full API, Kie Image API X, and Atlas Image API X nodes. They send prompts and optional reference images to supported providers and return images that can feed ordinary ComfyUI preview, processing, or save nodes. Available controls depend on the node and selected provider model; they may include size, aspect ratio, quality, seed, safety settings, and request timing.
API workflows require an appropriate provider credential. The supplied examples leave credentials blank, so configure them before queueing. For an object-editing example, see Anything Swap with NanoBanana. The Kie and Atlas example shows separate provider branches. Further details are in the NanoBanana guide and Kie and Atlas guide.
Process, compare, and replace image regions
- Image ProcessorX applies a saved edit recipe to one image or a two-image composition. You can adjust layers, opacity, curves, and other image settings, then choose which result the node outputs. Its Continue and Pause modes control whether the workflow waits for interactive editing. See the Image ProcessorX guide.
- Image Compare Edit X is an interactive endpoint for comparing two images, painting masks, making browser-side edits, and saving. It does not output an image to downstream nodes. See the comparison editor guide.
- DetailerX runs an ordered chain of available upscaling, detailer, and realism processors. Companion Preview and Masks nodes help you inspect intermediate images and detected regions. See the DetailerX guide.
- Anything Crop For Swap selects and prepares a region for editing; Anything Stitch places the edited crop back into the full image. Crop size, mask softness, and stitching or color-matching choices affect how naturally the replacement blends. See the Anything Swap guide.
Preview and save video
Preview Video X creates the MP4 needed for an in-node preview and passes the original video onward unchanged. Turn save output on when you also want that preview kept in the normal output directory.
Save Video X turns an image batch into video and can include audio. Its controls include frame rate, H.264/H.265/AV1 format, quality, pixel format, and output destination. Frame rate changes playback timing; quality and codec settings affect the encoded file. The node checks frame shape, codec settings, and FFmpeg availability when it runs. See the Save Video X guide.
Reuse workflows, prompts, and node groups
- XFlows organizes workflows with folders, tags, favorites, search, duplicate detection, and trash handling. See the XFlows guide.
- XPrompts stores searchable prompts and reusable preset blocks. Its Image Prompt Library can find embedded ComfyUI prompts in local images and show them alongside thumbnails. See the XPrompts guide and Image Prompt Library details.
- XNodes saves reusable nodes, selections, and native groups, including the connections contained within a saved selection. See the XNodes guide.
Organize the canvas
Every registered node’s right-click menu gains WorkflowX: Add to group and WorkflowX: Replace node.... The first places a node, or its selected set of nodes, into a native ComfyUI group. The second searches for another node and tries to retain compatible links, matching widget values, position, and appearance. It reports links it could not keep, so check the result before running the graph. Node-specific hidden properties are not guaranteed to transfer.
Additional audio and video tools
The included AuK audio nodes cover tasks such as generation and editing, transcription, instruction building, and chained voice cloning. See the chained-clone guide and AuK examples.
The H3 RefMod nodes support a reference-based video workflow with character, image, audio, and video reference tools. See the H3 RefMod guide and example workflows.
WorkflowX-Configurator Models
WorkflowX is not one image model, and the supplied README does not define a single set of built-in models to compare. Instead, different nodes help you select or work with model files and provider models:
- Diffusion models: Load Diffusion Model X switches among the model filenames you have added to its list. Only the active row is loaded for that node’s output.
- LoRAs: LoraX and Load Lora PickerX apply the LoRA files you select. Enabled rows, their order, and their separate model and CLIP strengths determine the applied stack.
- Prompting providers and profiles: Unified Autoprompter X offers target-model guidance and several remote or local provider choices. These choices shape how it prepares prompts; they do not automatically replace the diffusion model in your generation graph.
- Remote image models: The NanoBanana, Kie, and Atlas nodes expose supported provider-model choices. Their available settings vary by provider and model.
- Specialized assets: Some DetailerX, Anything Swap, AuK, and H3 RefMod workflows use their own selected models or supporting assets.
Choose a model according to the task and the node that uses it. For instance, switch the active diffusion-model row when testing two local generation setups; change a LoRA’s strength when you want more or less of its effect without changing the base model. The local generation and model-management example shows several of these controls together.
Troubleshooting WorkflowX-Configurator
An example opens, but its files or models are unavailable
- Check nodes with file or model selectors.
- Replace example placeholders with images, audio, video, models, and LoRAs available in your ComfyUI environment.
- Check for machine-specific paths, especially in workflows using Load Lora PickerX.
- Queue the graph only after its required resources are selected.
The examples intentionally contain no bundled media, credentials, or personal paths. Browse all examples from the example index.
A WorkflowX node or interface button is missing
- Restart ComfyUI after updating WorkflowX.
- Refresh the browser; use a hard refresh if a node’s button or another frontend control is still missing.
- Check the WorkflowX node menu and the XFlows, XPrompts, and XNodes interface areas.
- If a node remains missing, check whether its node ID appears in ComfyUI’s
/object_info.
Config SelectorX blocks queueing
- Confirm that there is only one initialized Config SelectorX on the root canvas. Nested selectors are ignored with a warning, while multiple initialized root selectors block queueing.
- Check the reported group names. Rename duplicates, including names that differ only by capitalization or surrounding spaces.
- Check matching Set keys in the winning priority tier. More than one eligible Set for the same key and family at that tier is an error.
- Review whether a group is Active, Bypassed, Muted, or Ignored in the selected configuration.
The Config SelectorX guide explains scope and value-resolution behavior in detail.
An API example does not produce an image
- Enter the required provider credential in the relevant API node.
- Review the selected provider model and its visible options.
- Check the node’s reported error for a timeout, provider failure, or safety rejection.
- For Kie or Atlas requests that remain pending, consult the Kie and Atlas guide for their retrieval path.
Do not share a workflow containing a credential.
Replacing a node leaves some links disconnected
- Read the replacement result notification for kept and skipped links.
- Inspect the new node’s inputs, outputs, and copied widget values.
- Reconnect any essential skipped links before queueing.
WorkflowX preserves compatible connections; it does not force incompatible socket types together.
Frequently asked questions
Does Config SelectorX replace native ComfyUI groups?
No. It manages native groups through its Scopes and Configs controls.
Can I use older configurator nodes in a new workflow?
Current workflows should use Config SelectorX. Earlier configurator nodes remain registered so old workflows can load; see the legacy migration guide.
Does sharing a workflow include its images, models, or credentials?
No. Advanced loader settings may travel with the workflow, but source files do not. Example API credentials are blank, and you should keep your own credentials out of shared graphs.
Does Image Compare Edit X send its edited image to another node?
No. It is a terminal interactive editor. Use an output-producing node, such as Image ProcessorX, when the result must continue through the graph.
Will turning off a LoRA row remove it from the applied stack?
Yes. LoraX applies enabled rows in their displayed order; empty rows are skipped.
Learn More about WorkflowX-Configurator
Start with the documentation index for feature-specific guides. For a hands-on introduction, open an example that matches your project:
- Configuration and routing for named setups and Set/Get values.
- Local generation and model management for model selection, LoRAs, prompting, unloading, and video output.
- Image loading, processing, and comparison for media and editing tools.
- Anything Swap with NanoBanana for crop, remote edit, and stitch.
- Kie and Atlas APIs for remote image branches.
- JsonX prompt toolchain for structured prompt creation.
- Advanced configured production for a larger configuration-driven graph.
WorkflowX is built for the ComfyUI ecosystem. The examples are starting points: inspect their nodes and connections, then select resources that are available to you before running them.
