WorkflowX Configurator

WorkflowX for ComfyUI

WorkflowX is a production toolkit for building, configuring, editing, organizing, and reusing ComfyUI workflows. It combines image and video nodes, model and LoRA controls, structured prompting, remote image APIs, reusable workflow libraries, scoped configuration, and canvas utilities in one package.

Short project brief: WorkflowX adds 42 active ComfyUI nodes plus XFlows, XPrompts, XNodes, and package-wide right-click utilities. Use it to build configurable workflows, manage LoRA stacks and VRAM, load, repair, convert, preview, and edit image/video/audio media, transform voices, create structured prompts, call supported image APIs, save video, store reusable graph fragments, group selections, and replace nodes without rebuilding compatible links.

Capabilities

| Area | What WorkflowX adds | Jump to | |---|---|---| | Media input | Thumbnail browsing, masks, visual video trimming/cropping, audio repair/conversion, dimensions, advanced geometry controls | Image and media loading | | Models | Ordered LoRA stacks and explicit model-component unloading | Model and LoRA management | | Prompting | Multi-backend prompt composition and structured JsonX tools | Prompting and JsonX | | APIs | Gemini/NanoBanana, Kie, and Atlas image generation/editing | Remote image APIs | | Image tools | Interactive compare/edit, deterministic processing, crop/stitch swap bridge | Image editing, processing, and swapping | | Output | H.264, H.265, and AV1 video encoding with optional audio | Video output | | Configuration | Config SelectorX, typed Set/Get values, and wildcard relay routing | Workflow configuration and routing | | Libraries | XFlows, XPrompts, and XNodes sidebar libraries | Workflow libraries | | Canvas | Native grouping and compatible node replacement from every node menu | Canvas right-click utilities |

Installation and settings

1. Place this repository in ComfyUI/custom_nodes/WorkflowX-Configurator.
2. Install optional dependencies required by the features you use.
3. Restart ComfyUI and refresh the browser.
4. Confirm WorkflowX appears in the node menu and XFlows, XPrompts, and XNodes appear in the interface.

WorkflowX settings let you enable or disable XFlows, XPrompts, and XNodes and export or import their local library data. Keep credentials and machine-specific paths out of workflows intended for sharing.

All downloadable examples are under examples/. They contain no bundled media, credentials, personal paths, prompts, or model-library names. Replace the neutral resource placeholders after loading.

Image and media loading

WorkflowX provides quick and advanced image loaders plus a visual video loader for workflows that need masks, dimensions, trimming, crop/resize controls, or serialized editor state.

Load ImageX

Node ID / category: WorkflowX_LoadImageX · WorkflowX/Image Loader

| Inputs and widgets | Type | Required | Behavior | |---|---|---:|---| | image | image-file combo | Yes | Selects an image from ComfyUI's input directory. | | Upload and Browse Thumbnails controls | UI | — | Uploads a file or opens the searchable thumbnail browser. |

| Outputs | Type | Behavior | |---|---|---| | image | IMAGE | Loaded RGB image batch. | | mask | MASK | Alpha-derived mask, or a default empty mask when no alpha is present. |

The backend validates the selected path against ComfyUI's input directory and participates in normal image-file change detection.

Load ImageX Adv

Node ID / category: WorkflowX_LoadImageXAdv · WorkflowX/Image Loader

| Inputs and state | Type | Required | Behavior | |---|---|---:|---| | image | image-file combo | Yes | Selects the source image. | | workflowx_state | STRING | UI-managed | Serialized advanced resize, crop, mask, and display state. | | Picker/editor buttons | UI | — | Open thumbnail browsing and advanced image controls. |

| Outputs | Type | Behavior | |---|---|---| | image | IMAGE | Processed image. | | mask | MASK | Effective mask. | | inverted_mask | MASK | Inverse of the effective mask. | | width | INT | Result width. | | height | INT | Result height. |

The serialized state travels with the workflow; source files do not. See the advanced loader guide.

Load VideoX Adv

Node ID / category: WorkflowX_LoadVideoXAdv · WorkflowX/Video

| Inputs and state | Type | Required | Behavior | |---|---|---:|---| | video | video-file combo | Yes | Recursively selects or uploads a video in ComfyUI's input directory. | | seconds | FLOAT | No | When connected, locks the visual trim range to that duration while allowing the range to slide. | | workflowx_state | STRING | UI-managed | Versioned trim, crop, resize, snap, resample, and spatial-padding state. |

| Outputs | Type | Behavior | |---|---|---| | video | VIDEO | Native lazy trim when possible; otherwise a rebuilt trimmed/cropped video preserving source FPS, bit depth, color space, and available audio. | | video_frames | IMAGE | The selected frame batch with the same crop and geometry transform. | | audio | AUDIO | Trimmed source audio, or None for silent sources. | | width | INT | Final output width. | | height | INT | Final output height. |

The node face provides the full Load ImageX Adv geometry toolset and reflows when mode-specific controls appear. ComfyUI's native video player is the only node-face preview, with a transparent crop overlay attached directly to it; the video picture is reserved for crop interaction and playback starts from the native play control. Open Timeline creates a graph-following, canvas-layer-aware floating player with a thumbnail filmstrip, large in/out handles, draggable selection, separate playhead scrubber, editable selection seconds, selection playback, frame/time readouts, lock, minimize, resize, and fullscreen controls. Media outputs are processed only when connected, so width/height-only graphs remain metadata-only. See the Load VideoX Adv guide.

Voice ChangerX

Node ID / category: WorkflowX_VoiceChangerX · WorkflowX/Audio

Transforms a single spoken voice with independent pitch, formant, brightness, breathiness, pitch-variation, and gain controls. Every control has a slider and numeric box, with fine decimal adjustment and Reset all. The only sockets are source audio, optional reference audio, and the audio output. Match Reference compares the connected voices and fills the editable sliders with approximate acoustic matching settings. It queues only the source/reference dependencies and this node. Connect the output to ComfyUI Preview Audio and run the workflow to audition. Normal runs use the sliders without rematching. Neutral settings return the original source exactly; transformed output preserves sample rate, duration, batch and channel counts.

Uses local CPU processing through PyWORLD; no model downloads or online services. See the Voice ChangerX guide for ranges, installation, and tuning.

Load AudioX

Node ID / category: WorkflowX_LoadAudioX · WorkflowX/Audio

| Inputs and state | Type | Required | Behavior | |---|---|---:|---| | seconds | FLOAT | No | Locks the processed output to a wired duration, using silence padding or looping when needed. | | LoadAudioXState | STRING | UI-managed | Versioned file, trim, conversion, recovery, effect, and destination state. |

| Outputs | Type | Behavior | |---|---|---| | audio | AUDIO | Trimmed and processed float32 PCM master for normal ComfyUI audio nodes. | | converted_path | STRING | Absolute path to the MP3 or WAV encoded from that same master. |

The compact face keeps file selection, waveform playback, draggable trim, duration mode, output format, depth/bitrate, channel layout, sample rate, and Repair controls directly on the node. The gear popup holds bounded recovery diagnostics, cleanup presets, gain/normalization/limiting, silence trim, fades, and permanent-save settings. It accepts common audio containers and the first audio stream of video, tolerantly salvages isolated damaged MP3 frames, and applies a narrowly verified recovery for audio-only AAC M4A files whose stsz table is truncated but whose metadata and payload prove an exact reconstruction. The source file is never modified. See the Load AudioX guide.

Working example

Download: image/video loading, processing, and comparison — select two local images and optionally a video, then inspect masks, processed images, a trimmed/cropped frame batch, the native video preview, and the terminal comparison.

Model and LoRA management

Load Diffusion Model X keeps preferred diffusion models visible on the canvas and switches between them with a single active selection. LoraX applies a visible, ordered LoRA stack. Unload Models By Type adds an execution dependency for releasing selected model components after the graph stage you choose.

Load Diffusion Model X

Node ID / category: KVGC_LoadDiffusionModelX · WorkflowX/Loaders

| Inputs | Type | Required | Behavior | |---|---|---:|---| | weight_dtype | combo | Yes | Uses the choices and behavior exposed by ComfyUI's native Load Diffusion Model node. | | Diffusion-model rows | UI-managed | Yes | Stores multiple preferred model filenames; exactly one row is active. |

| Outputs | Type | Behavior | |---|---|---| | MODEL | MODEL | The active row loaded through ComfyUI's native UNETLoader. |

Controls: add, replace, remove, search, refresh, inspect model details, and switch the active model with radio-style toggles. The picker uses LoRA Manager metadata and previews when available, while remaining fully usable from ComfyUI's diffusion-model folders alone. The shared weight_dtype control applies to whichever row is active. Load Diffusion Model X guide

LoraX

Node ID / category: KVGC_LoraX · WorkflowX/Loaders

| Inputs | Type | Required | Behavior | |---|---|---:|---| | model | MODEL | Yes | Base diffusion model. | | clip | CLIP | No | Optional text encoder to receive LoRA patches. | | LoRA rows | UI-managed | No | Searchable LoRA selection plus model/CLIP strengths. |

| Outputs | Type | Behavior | |---|---|---| | MODEL | MODEL | Model with enabled LoRAs applied in displayed order. | | CLIP | CLIP | Patched CLIP when supplied. | | trigger_words | STRING | Trigger words gathered from enabled rows. | | loaded_loras | STRING | Human-readable audit of the applied stack. |

Controls: add, remove, reorder, search/select, enable/disable, and refresh LoRA rows. Empty rows are skipped; unavailable filenames are reported instead of silently substituted. LoraX guide

Unload Models By Type

Node ID / category: KVGC_UnloadModelsByType · WorkflowX/VRAM

| Inputs | Type | Required | Behavior | |---|---|---:|---| | model_type | combo | Yes | Text Encoder, Diffusion Model / UNet, VAE, CLIP Vision, Other, or All loaded models. | | device_scope | combo | Yes | Current device or all devices. | | empty_cache | BOOLEAN | Yes | Requests device-cache cleanup after unloading. | | trigger | wildcard | No | Establishes when unloading runs. | | model, clip, vae, conditioning | matching types | No | Optional typed passthrough values. |

| Outputs | Type | Behavior | |---|---|---| | trigger | wildcard | Preserves the execution dependency. | | model, clip, vae, conditioning | matching types | Typed passthroughs. | | status | STRING | Reports what was requested and released. |

Working example

Download: local generation and model management — Load Diffusion Model X selects the active model, Load ImageX guides Unified Autoprompter X, LoraX participates in sampling, and the generated batch triggers model unloading before Save Video X.

Prompting and JsonX

WorkflowX supports natural or structured prompting from images and text, then provides a connected JsonX toolchain for generating, inspecting, importing, and randomizing structured prompt documents.

Unified Autoprompter X

Node ID / category: UnifiedAutoprompterX · WorkflowX/Prompting

| Inputs and widgets | Type | Required | Behavior | |---|---|---:|---| | target_model, generation_type, prompt_format | combo | Yes | Select target-model conventions, a downstream generation path, and natural, tag, or JSON formatting. | | nsfw_enabled, negative_enabled, enable_bbox_json_input, enable_text_input | BOOLEAN | Yes | Enable optional path rules and prompt channels. | | refresh_vram, disable_color_palette | BOOLEAN | Yes | Control model refresh and palette output. | | generated_positive, generated_negative, final_prompt | STRING | UI-managed | Stores generated and user-edited prompt text. | | image, dynamic image_N | IMAGE | No | Visual references. | | bbox_json, raw_prompt_text | STRING | No | Structured layout or source text. | | ui_state | STRING | UI-managed | Serialized frontend composition state. |

| Outputs | Type | Behavior | |---|---|---| | prompt | STRING | Final selected-format prompt. | | positive | STRING | Positive prompt channel. | | negative | STRING | Negative prompt channel when enabled. |

Controls: one Provider dropdown for Gemini, Grok API, DeepSeek API, OpenAI Compatible, LM Studio, Unsloth Studio, Ollama, or Local GGUF; profile-aware Generation type; Prompt instructions; Detail level; NSFW/negative controls; dynamic Model settings; bounding-box tools; Profile settings; Generate and Cancel. Provider credentials and tuning remain browser-local. See the complete Autoprompter guide.

Connected images are optional prompt-authoring evidence and never select the downstream generation path. Image-capable paths switch between with-reference and inferred-reference instructions; text-only paths can inspect connected media and combine visible evidence with the user's description without exposing reference commentary downstream. Standard-profile instructions are Markdown-backed: packaged defaults live under unified_autoprompter/reference/original, while the editable runtime copy lives under reference/current_use. Profile Settings edits the exact Common, selected Generation Path, Reference Usage, Output Contract, and Image/Video NSFW files without reformatting them. JsonX has an independently managed reference/original/JsonX and reference/current_use/JsonX subtree with exact editors for paths, conditions, user templates, context templates, presets, and stage contracts. Preview and dispatch share their respective backend payload builders and keep routing metadata out of model-facing text.

JsonX visual and template tools

| Node | Internal ID | Inputs | Outputs | Principal controls | |---|---|---|---|---| | JsonX - Visual Builder | FluxVisualJsonBuilder | Optional prompt_json (STRING) | prompt_json (STRING) | Open builder, schema fields, presets, apply, reset | | JsonX - Prompt Template Importer | AFJPromptTemplateImporter | Optional template_name, source_prompt_json, import_report (STRING) | template_payload_json (STRING) | Open importer, analyze/import, apply, clear | | JsonX - Template Randomizer | FluxTemplateRandomizer | Optional template_name, randomize_rules, randomize_rules_help (STRING), seed (INT) | prompt_json, run_log (STRING) | Template picker, rule editor, seeded randomization |

The builder normalizes JSON before output; the importer reports accepted and skipped fields; the randomizer applies reproducible field choices and emits an audit log. See the JsonX user guide.

Working example

Download: JsonX prompt toolchain — the first lane connects Unified Autoprompter X with its JsonX profile → Visual Builder → Template Importer; the second connects Template Randomizer → a second builder for inspection.

Remote image APIs

API nodes are deliberately credential-free in shared workflows. Their outputs use ordinary IMAGE sockets, so remote generation can feed the same preview, processing, and save nodes as local generation.

NanoBanana Full API

Node ID / category: NanoBanana_Gemini_2_5_Flash_V2 · WorkflowX/API

| Input group | Exact inputs | Type / behavior | |---|---|---| | Authentication and model | api_key, model_version | Private string and supported Gemini image-model combo. | | Prompting | prompt, system_prompt | Generation/edit instructions. | | Sampling and output | aspect_ratio, seed, temperature, top_p, candidate_count, resolution, timeout_seconds | Request and output controls. | | Safety | safety_harassment, safety_hate_speech, safety_sexual, safety_dangerous | Per-category blocking policy. | | Editing/thinking | edit_mode_enabled, show_thoughts, thinking_level | Edit and reasoning controls. | | Optional media | mask, image_1 … image_5 | Mask and reference images. |

| Outputs | Type | Behavior | |---|---|---| | image_batch | IMAGE | Returned candidate images. | | text_output | STRING | Provider text or response details. |

Empty credentials, provider errors, timeouts, and safety failures are surfaced. See the NanoBanana API guide.

Kie and Atlas Image API X

Node IDs / category: WorkflowX_KieImageAPI, WorkflowX_AtlasImageAPI · WorkflowX/API

| Input group | Exact shared inputs | Type / behavior | |---|---|---| | Request | api_key, model, prompt, aspect_ratio, image_size | Provider credential, model, prompt, and dimensions. | | Polling | timeout_seconds, poll_interval_seconds, retrieval_mode | Submit/poll or force-retrieve behavior. | | Quality | quality, guidance_scale, num_inference_steps, media_resolution, input_fidelity, enable_pro | Provider/model-aware quality controls. | | Safety and provenance | nsfw_checker, enable_safety_checker, watermark, show_payload | Safety, watermark, and payload visibility. | | Generation options | thinking_mode, seed_enabled, seed, enable_sequential, output_format, enable_web_search, enable_image_search | Optional provider features. | | Custom dimensions | custom_size_enabled, custom_size_auto, custom_width, custom_height, reference_max_edge | Custom output/reference sizing. | | References | Dynamic image_1 … image_14 | Frontend adds image sockets as references are connected. |

| Node | Outputs | Provider-specific behavior | |---|---|---| | WorkflowX_KieImageAPI | image (IMAGE) | Kie submission, polling, pending-task, and retrieval path. | | WorkflowX_AtlasImageAPI | image (IMAGE) | Atlas submission and retrieval path with its supported model/size catalog. |

The frontend changes visible controls according to the selected model. Pending records allow later force retrieval. See the Kie and Atlas guide.

Working examples

• Anything Swap with NanoBanana connects a cropped region, crop mask, and swap prompt to the API before stitching the returned image.
• Kie and Atlas APIs uses separate provider branches with an optional shared reference image and output previews.

Image editing, processing, and swapping

Image ProcessorX

Node ID / category: WorkflowX_ImageProcessorX · WorkflowX/Image Compare

| Inputs and state | Type | Required | Behavior | |---|---|---:|---| | image1 | IMAGE | Yes | Primary image. | | image2 | IMAGE | No | Optional second layer/image. | | operation_mode | Continue / Pause | Yes | Executes immediately or pauses for interactive editing. | | output_image | O1 / O2 / O3 | Yes | Selects source one, source two, or composed output. | | processor_state | STRING | UI-managed | Versioned, normalized edit recipe. |

| Outputs | Type | Behavior | |---|---|---| | image | IMAGE | Selected or composed result. |

Controls: open editor, apply/continue, reset, layer order, opacity, adjustments, curves, presets, and save/session actions. State is validated server-side so a reloaded workflow renders deterministically. Image ProcessorX guide

Image Compare Edit X

Node ID / category: KVGC_ImageCompareEditX · WorkflowX/Image Compare

| Inputs | Type | Required | Behavior | |---|---|---:|---| | image1 | IMAGE | Yes | First comparison image. | | image2 | IMAGE | Yes | Second comparison image. |

| Outputs | Type | Behavior | |---|---|---| | None | — | Terminal interactive output node; browser-side edits do not emit a downstream tensor. |

Controls: comparison divider, preview navigation, layers, masks, blend mask, brush, adjustments, curves, reset, copy, and save. Editor guide

Anything Swap bridge

Anything Crop For Swap

Node ID / category: AnythingCropForSwap · WorkflowX/Anything Swap

| Input group | Exact inputs | Behavior | |---|---|---| | Source | image; optional mask, caption | Source image plus optional precomputed mask/caption. | | Detection | use_sam3, sam3_prompt, sam3_checkpoint, threshold, refine_iterations, keep_model_loaded, select_mode, object_index | Optional SAM3 object selection. | | Geometry | expand_factor, expand_pixels, force_square, padding, edge_handling, resize_mode, target_size, upscale_method, downscale_method | Crop and resize policy. | | Mask/prompt | mask_grow, mask_blur, swap_prompt | Final crop mask and downstream edit instruction. |

| Outputs | Type | Behavior | |---|---|---| | crop | IMAGE | Selected crop. | | crop_mask | MASK | Crop-space mask. | | swap_prompt | STRING | Prompt passthrough. | | source_masked | IMAGE | Source with selected region visualized/masked. | | stitch | SWAP_STITCH | Geometry payload consumed by Anything Stitch. | | detected | BOOLEAN | Whether a usable region was found. |

Anything Stitch

Node ID / category: AnythingStitch · WorkflowX/Anything Swap

| Inputs | Type | Required | Behavior | |---|---|---:|---| | stitch | SWAP_STITCH | Yes | Crop geometry from Anything Crop. | | swapped | IMAGE | Yes | Edited/replaced crop. | | mask_override | MASK | No | Optional replacement mask. | | size_mismatch, mask_mode, feather, color_match, color_match_method, color_match_strength | widgets | Yes | Resize, mask, feather, and color-matching policy. |

| Outputs | Type | Behavior | |---|---|---| | image | IMAGE | Stitched full image. | | changed_mask | MASK | Effective changed region. |

See the Anything Swap guide.

Working examples

• Image loading, processing, and comparison
• Anything Swap with NanoBanana

Video output

Preview Video X

Node ID / category: WorkflowX_PreviewVideoX · WorkflowX/Video

| Inputs | Type | Required | Behavior | |---|---|---:|---| | video | VIDEO | Yes | Native ComfyUI video displayed as an animated in-node player. | | save_output | BOOLEAN | Yes | Off writes only the temporary preview; on also makes the encoded MP4 a persistent output. | | filename_prefix | STRING | Yes | Relative preview/output name prefix. |

| Outputs | Type | Behavior | |---|---|---| | video | VIDEO | Passes the original video through unchanged. | | filenames | VHS_FILENAMES | Describes the temporary or saved preview file. |

The node always materializes the minimum MP4 needed for ComfyUI's player. With save output off that file lives in the temporary directory; with it on the same preview is written to the normal output directory. H.264 previews of odd-sized videos receive at most one black pixel on the right or bottom to satisfy the codec; the pass-through VIDEO remains unchanged.

Save Video X

Node ID / category: WorkflowX_SaveVideoX · WorkflowX/Video

| Inputs | Type | Required | Behavior | |---|---|---:|---| | images | IMAGE | Yes | Image batch encoded as frames. | | audio | AUDIO | No | Optional audio stream. | | vae | VAE | No | Optional compatibility input. | | frame_rate, filename_prefix, format | widgets | Yes | Timing, relative output prefix, and H.264/H.265/AV1 container. | | crf_quality, pixel_format, color_range | widgets | Yes | Codec quality and pixel/color policy. | | save_metadata_preview_image, audio_bitrate, audio_filters, save_output | widgets | Yes | Metadata preview, audio encoding, filters, and preview/output destination. |

| Outputs | Type | Behavior | |---|---|---| | Filenames | VHS_FILENAMES | Encoded file descriptors compatible with video-preview consumers. |

Frame shape, codec settings, and FFmpeg availability are validated at execution. Save Video X guide

Working example

The local generation and model management workflow turns a generated batch into video and uses the batch as the trigger for model unloading before encoding.

Workflow configuration and routing

Config SelectorX is WorkflowX's integrated configuration system. Its Scopes and Configs buttons manage native ComfyUI groups directly; the previous separate configurator nodes are not part of current workflows.

Config SelectorX

Node ID / category: KVGC_ConfigSelectorX · WorkflowX/Workflow Config

| Inputs and state | Type | Required | Behavior | |---|---|---:|---| | selected_config | STRING | Yes | Active named configuration. | | console_output | no / yes | Yes | Enables resolution diagnostics. | | selectorx_state | STRING | UI-managed | Versioned scopes, configurations, group modes, mute state, and bypass state. |

| Outputs | Type | Behavior | |---|---|---| | None | — | Configuration controller; it changes scoped node participation and Get resolution. |

Controls: Scopes assigns each native group to configuration, selector-mute, selector-bypass, or ignore behavior. Configs creates named configurations and assigns Active, Bypass, Mute, or Ignore per controlled group. State is stored in workflow JSON and validated during execution. Config SelectorX guide

Typed Set/Get values

Set nodes publish a keyed value from their scoped group and intentionally have no output. Get nodes resolve the selected candidate and expose it to the execution graph. resolved_value, resolved_config, and resolved_digest are visible but UI-managed provenance fields on every typed Get node.

| Pair | Internal IDs | Set contract | Get output | |---|---|---|---| | Integer | KVGC_SetInt, KVGC_GetInt | key: STRING, value: INT; no outputs | int: INT | | Float | KVGC_SetFloat, KVGC_GetFloat | key: STRING, value: FLOAT; no outputs | float: FLOAT | | String | KVGC_SetString, KVGC_GetString | key: STRING, value: STRING; no outputs | string: STRING | | Multiline text | KVGC_SetText, KVGC_GetText | key: STRING, value: STRING; no outputs | text: STRING | | Boolean | KVGC_SetBoolean, KVGC_GetBoolean | key: STRING, value: BOOLEAN; no outputs | boolean: BOOLEAN | | Sampler | KVGC_SetSampler, KVGC_GetSampler | key: STRING, sampler combo value; no outputs | sampler_name: live ComfyUI sampler combo | | Scheduler | KVGC_SetScheduler, KVGC_GetScheduler | key: STRING, scheduler combo value; no outputs | scheduler: live ComfyUI scheduler combo |

Sampler and scheduler values are validated against the running ComfyUI installation. Typed resolution follows native group containment and the selected Config SelectorX state.

Relay routing

| Node | Internal ID | Inputs | Outputs | Behavior | |---|---|---|---|---| | Set Relay | KVGC_SetRelay | value (wildcard), key (STRING) | value (wildcard passthrough) | Publishes a scoped object while optionally keeping it on the local execution path. | | Get Relay | KVGC_GetRelay | key (STRING), optional fallback value (wildcard) | value (wildcard) | Resolves the selected routed object or returns the connected fallback. |

Dimension routing

Set/Get Dimensions relays a keyed width-and-height pair through the same configuration-aware selection mechanism. The pair has no mute controls.

| Node | Internal ID | Inputs | Outputs | Behavior | |---|---|---|---|---| | Set Dimensions | KVGC_SetDimensions | key (STRING), width (INT), height (INT) | width (INT), height (INT) | Publishes and locally passes through both dimensions. Values default to 1024 and are constrained to 1–16384. | | Get Dimensions | KVGC_GetDimensions | key (STRING), optional direct width and height inputs | width (INT), height (INT) | Resolves both values from the selected Set Dimensions or uses both directly connected fallbacks. |

Reference routing with independent mute controls

Set/Get Reference routes images, audio, video, or any other connected ComfyUI object through the same keyed, configuration-aware selection used by Relay. Unlike Relay, each Get Reference can apply ComfyUI's native Mute mode to its own branch while the prompt is serialized.

| Node | Internal ID | Inputs | Output | Behavior | |---|---|---|---|---| | Set Reference | KVGC_SetReference | value (wildcard), key (STRING), mute (BOOLEAN, default false) | value (wildcard passthrough) | Always passes its local value through. When mute is true, every matching Get Reference is omitted from the API prompt using native ComfyUI mute semantics. | | Get Reference | KVGC_GetReference | key (STRING), mute (BOOLEAN, default false), optional fallback value (wildcard) | value (wildcard) | Resolves the selected Set Reference. When its independent mute is true, that Get node is omitted from the API prompt exactly like manually muting the node. |

Set Reference mute applies only to Get Reference nodes with the same key. Muting one Get Reference does not affect its siblings, and the Set Reference output remains usable directly in either mute state.

Working example

Download: configuration and routing — Draft and Final native groups provide typed values and alternative model relays; resolved values drive real sampler and text-encoding inputs.

The larger advanced configured production workflow shows the same relay pattern inside a model → LoRA → sampler → cleanup → video pipeline.

Workflow libraries

These are sidebar features rather than executable nodes, so they are demonstrated as interface workflows instead of artificial JSON nodes.

XFlows

XFlows organizes ComfyUI workflows with folders, list/card views, tags, favorites, search, move tools, duplicate detection, and trash handling without changing the workflow JSON format.

XFlows guide

XPrompts

XPrompts stores searchable prompts and reusable preset blocks for insertion and editing. It complements structured JsonX workflows without forcing prompt text into a workflow graph.

XPrompts guide

XNodes

XNodes saves reusable nodes, selections, and native groups. Group snippets preserve selected ComfyUI groups, group geometry and styling, contained nodes, widget values, and links whose endpoints are both inside the saved selection.

XNodes guide

Canvas right-click utilities

WorkflowX adds two options to every registered node's context menu.

WorkflowX: Add to group

• If the clicked node is part of the current selection, all selected nodes are placed in one native ComfyUI group.
• If the clicked node is not selected, only that node is grouped.
• Group bounds use ComfyUI's Group Selected Nodes Padding setting, with a safe default when the setting is unavailable.

WorkflowX: Replace node...

The searchable dialog lists every registered node except the current class and can be filtered by display name, internal class, or category.

On replacement, WorkflowX preserves position, colors, mode, flags, a custom title, and a size large enough for the replacement. Matching widget names are copied. Input and output slots are matched by name and compatible type first, then by compatible type. Wildcard sockets are supported.

Incompatible links are skipped rather than forced, and a result notification reports kept links, skipped links, and copied widget values. Arbitrary node-specific hidden properties are not guaranteed to transfer.

Examples

| Workflow | Demonstrates | Execution | |---|---|---| | Configuration and routing | Config SelectorX, every typed Set/Get pair, Relay, real consumers | Model-dependent | | Local generation and model management | Load Diffusion Model X, Load ImageX, Unified Autoprompter, LoraX, unload ordering, Save Video X | Model/resource-dependent | | Image loading, processing, and comparison | Both loaders, masks, Image ProcessorX, Image Compare Edit X | Local images required | | Anything Swap with NanoBanana | Advanced loader, crop, remote edit, stitch, previews | API-dependent | | Kie and Atlas APIs | Separate provider branches with previews | API-dependent | | JsonX prompt toolchain | Unified JsonX, builder, importer, randomizer | Partly local/profile-dependent | | Advanced configured production | Sanitized production pattern using configuration, relay, LoRA, cleanup, and video | Model-dependent |

Acknowledgements

WorkflowX Configurator is built for the ComfyUI ecosystem and benefits from the ideas, conventions, and open work shared across its custom-node community. The project as a whole, not only the nodes named below, reflects that wider ecosystem. Particular thanks go to:

• ComfyUI-Pixaroma for UI inspiration and the adjustment behavior adapted for Image Compare Edit X.
• ComfyUI-Curve for the curve-editor interaction model adapted for Image Compare Edit X.
• ComfyUI-Lora-Manager for the metadata, preview, and model-library APIs with which LoraX interoperates, and for helping establish the visual LoRA-management patterns LoraX develops further.
• llama.cpp, Ollama, and the prompting guidance published across supported model ecosystems, which informed Unified Autoprompter X's local and provider-backed prompt workflow.
• The original Anything Swap Bridge implementation and AnythingCropForSwap / AnythingStitch contract, which WorkflowX preserves while extending the workflow with model-agnostic editing, stronger geometry and mask handling, and native ComfyUI SAM3 support.
• ComfyUI NanoBanana Full API for the original NanoBanana node that WorkflowX updates while preserving workflow compatibility.
• The bundled AFJ project for its visual JSON prompting foundation, and GemMobi for the canonical model contracts used by the Kie and Atlas image API nodes.

Some WorkflowX nodes preserve an established workflow contract while expanding, enhancing, or adapting its functionality for different use cases and a more unified experience. Others combine familiar interaction patterns with new implementations or interoperate with adjacent projects. These acknowledgements recognize those foundations; the resulting WorkflowX nodes may differ substantially in interface, scope, and behavior.

Migration, compatibility, and troubleshooting

• Current workflows should use Config SelectorX. The four earlier configurator nodes remain registered only so old workflows can load; see legacy migration.
• Example resource names are placeholders. Select local images, checkpoints, LoRAs, masks, audio, and codecs after loading.
• API examples have blank credentials and should not be queued until credentials and provider settings are configured.
• If a node is missing, restart ComfyUI after updating WorkflowX and confirm the node ID appears in /object_info.
• If a UI button is missing, hard-refresh the browser so the current frontend extensions load.
• For detailed feature guides, use the documentation index.

WorkflowX does not rename active node IDs or rewrite ComfyUI's workflow serialization format.