The AI assistant that lives inside ComfyUI
— now in free beta

Every node gets a name you can point at

WayFinder gives each node a short, stable tag such as (KSM.01) at the start of its title. Replies, change summaries, errors, run progress and notes all point at exactly the node they mean — no more "which KSampler?".

• Tags in the chat are clickable: one click centers the node on the canvas
• Type /ref to pick a node and drop its tag into your message
• Tags are saved with the workflow and survive reloads, renames and undo
• Documentation notes end with an index of every node and connection

Subgraphs and a reusable module library

Package part of a workflow as a subgraph, save it to your encrypted library, and reuse it in other workflows through the WayFinder Subgraph node. In Architect mode, each stage it builds lands on your canvas as its own subgraph, tagged so you can find it again.

A setup wizard that proves it works

Pick your AI and hit Test connection for a real round trip before you commit — it won't let you leave with a dead key. Then switch on only the extras you want: audio transcription, long-term memory, MCP agent tools, video analysis or a ComfyUI GPU host. Skip any step, re-run it from Settings, and never hand-edit a config file.

Untangle any workflow you downloaded

Open someone's messy graph and WayFinder lays it out so you can actually read it — clean columns, labeled group boxes with color-coded nodes by function, fewer crossed wires — then explains what each part does and notes what it needs to run. The auto-group tool draws boxes around same-function clusters (loaders, samplers, VAE, output…) while keeping any groups you already made. Add force to wipe everything and regroup from scratch.

Run it and watch — WayFinder catches the errors

It runs your image, tracks the progress, and reads the result. When something fails, it translates the cryptic error into plain words and tells you how to fix it — instead of leaving you staring at a red node.

Nothing you can't undo

Every change WayFinder makes is reversible with one click, and it always asks before doing anything risky — installing, downloading, or restructuring. You can also pre-authorize trusted action categories (like "always run workflows on this canvas") so routine operations never stop for a card.

It checks before it runs

WayFinder flags missing models and unset settings before you waste a render, predicts whether your GPU can handle the job so you avoid out-of-memory crashes, and checks that your prompt actually matches the workflow's model type — image-edit workflows expect a short edit instruction, not a full character description — before tweaking values. It also works out which model family your workflow is built on and warns when a LoRA, VAE or ControlNet looks like it belongs to a different one. For video, /vram scales its estimate by the clip's frame count.

Make 50 versions without rebuilding 50 times

Set up your variations once — colors, poses, prompts — on a single node, and WayFinder runs the whole batch in one approval and labels every result. You can also pick an image from your personal asset library as a variable value. Because the settings live on the node, every saved image remembers exactly how it was made, so next month is a reload, not a rebuild.

• Navigate between sets with ◀/▶ buttons directly on the canvas node
• Run all named variable sets in one approval — or pick exactly which ones to queue
• Browse your other workflows and import any of their variable sets — no copy-pasting needed
• Pick a library image directly as an IMAGE-type variable value
• Rename any variable key — WayFinder updates it everywhere (every set, the type map, display order, and seed mode) in one step, so nothing gets out of sync and canvas wiring stays intact
• Change a variable's type safely — WayFinder checks every existing value first and reports any incompatibilities before committing
• Version history — every destructive edit takes a snapshot of your sets first (the last 20 are kept), so you can list earlier versions and restore one
• AUDIO and VIDEO variables take clips from your asset library, just like IMAGE variables
• Use the SEED type for KSampler seed variables — set it to random, fixed, or increment and WayFinder resolves the right seed value on every run automatically, without you editing values between runs
• Duplicate a workflow with /duplicate — WayFinder writes a copy under a new name while you keep working in the original, and automatically forks its variable sets so the copy is fully independent

Match a pose or a reference in one sentence

Drop in a reference image from your asset library and say "use this pose." WayFinder wires up the pose, depth, or composition control for you, adds sharper upscaling, and fixes faces — the setup that normally takes a whole tutorial to get right.

Compare versions side by side — drag a live wipe slider

Run a settings sweep, then ask WayFinder to "open the compare view" — or type /compare in the chat composer for an instant overlay with no agent turn needed. A fullscreen overlay shows a thumbnail strip of all your results. Drag the ⇆ wipe slider left and right (or top to bottom — switch the orientation in one click) to reveal any two results side by side. Scroll to zoom in up to 6× — the zoom anchors to your cursor so the detail you care about stays in frame — then drag to pan. Both layers stay perfectly aligned so nothing shifts as you move the slider. No images go through the AI chat; everything loads instantly. Inline chat images also have a one-click zoom, download, and "add to library" action right on the hover bar.

• Thumbnail strip — click any result to assign it to slot A or B
• Draggable wipe slider with vertical or horizontal orientation
• Synchronized zoom (up to 6×, cursor-anchored) and pan across both layers
• ✦ Analyze A vs B — clicking Analyze submits just those two results to WayFinder for a written verdict; the result appears directly in chat
• Analyze all N — send the full sweep to the AI for a set-level review
• Each analysis result includes an Ask a follow-up button — click it to open a dialog and drill in with another question without leaving the chat
• /analyze slash command — skip the overlay and get a written AI evaluation of your most recent outputs in one keystroke

See what's eating your disk — and clean up safely

The model manager shows how much space your models take, which saved workflows and templates use each one, and which aren't used by anything. Before anything is deleted you see the exact files, the space you'd get back, and every workflow that would break — and nothing is removed until you confirm.

Learn by doing, inside your own canvas

A patient tutor (Comfy U mode) builds the example with you and explains every choice — no separate course, no tab-switching. It even pulls its own reference images so you're not hunting for files. And it adapts its explanations to your level: deep dives where you're a beginner, short-cuts where you already know the terrain. WayFinder tracks your skill level across seven ComfyUI topic areas and updates its language register automatically as it learns from you — so explanations always land at the right depth, whether you're in a lesson or just asking questions. You come away knowing ComfyUI, not just holding a finished workflow.

Plan a big project, approve it step by step

For a character sheet or a full product line, WayFinder proposes a staged plan with a risk register and a decision-point map, you approve each stage, and it builds them one block at a time. Save the plan to your library and re-run it later — a studio-grade pipeline that comes back, not a one-off. See how it scales →

The WayFinder Essentials course and lesson packs

New to WayFinder? In Comfy U, ask "teach me how to use WayFinder" for ten hands-on lessons: the sidebar, modes, slash commands and approvals, the WayFinder nodes, the asset library, compare and analyze, memory, permissions and Architect. Lesson packs on image foundations, control and editing, Flux, upscaling, and video, audio and 3D track your progress, and each lesson opens its workflow in a new tab so yours stays as it is.

Find the right model or style — then add it in a click

Search for models, styles (LoRAs), and add-ons across Hugging Face, Civitai, and the node registry — results ranked by popularity and freshness so you pick something well-maintained. WayFinder even tells you what a model is for before you use it.