Utilities¶

Workflow plumbing: counters, variables, timers, inspection, JSON loading, and small math helpers.

At a glance¶

Node	Summary
Better Int	An integer that increments or randomizes itself before or after each run, with the widget kept in sync.
Gated Increment	An integer that only changes once every N runs — the counter survives restarts.
Property	A named variable: wire a value in to set it, leave it unwired to get it — teleport values across a graph.
Property Clear	Clears stored properties by scope and/or key.
Property List	Lists all stored properties with value summaries.
Random Select	Randomly forwards one of up to 5 connected inputs.
Gate Any	Pass any value through, or block the branch with a boolean.
Float Lerp	Unclamped linear interpolation between two floats.
Timer Start / Timer Stop	Measure wall-clock time between two points in a graph.
Print	Prints a custom message to the console when it executes.
Inspect Tensor	Type-aware inspection of any value, with pass-through.
Load JSON File	Loads a JSON file from the input directory via a dropdown.
Load JSON File Path	Loads a JSON file from an arbitrary path.
Token Counter	Counts a prompt's tokens against the model's budget.
Conditioning Token Count	Reads the sequence length straight from a conditioning tensor.

Counters & values¶

Better Int¶

An integer constant that can change its own value — by a fixed step or by randomizing — before or after each run, with the on-node widget kept in sync.

Unlike the built-in control_after_generate (which only mutates the widget after execution), the *_before modes apply the change first, so the output reflects the new value on the current run. After each run the widget shows the current number, so you can pause a batch, read it, and resume.

Parameter	Type	Default	Description
`value`	INT	0	The current integer (base for increments).
`mode`	choice	`fixed`	`fixed`, `increment_before`, `increment_after`, `random_before`, `random_after`.
`step`	INT	1	Amount added per increment.
`debug`	BOOLEAN	false	Log mode/input/output details.

Outputs: INT — the value emitted this run.

Random modes draw from 0–4294967295 (seed-sized). In any non-fixed mode the node re-executes on every queue.

Gated Increment¶

An integer that only changes once every N runs — hold a seed or value stable for a few generations, then advance.

Each execution advances an internal counter; when it reaches rate the value increments (or randomizes) and the counter resets. The counter is persisted to disk per node, so it survives ComfyUI restarts, and the node's title shows the live count, e.g. Gated Increment (CCN) [3 / 5].

Parameter	Type	Default	Description
`value`	INT	0	Current integer / base for increment.
`rate`	INT	5	Number of runs between each value change.
`mode`	choice	`increment`	`increment` adds `step`; `random` picks a new seed-sized value.
`timing`	choice	`before`	Whether the changed value is emitted this run (`before`) or next (`after`).
`step`	INT	1	Amount added in increment mode.
`reset_counter`	BOOLEAN	false	Reset the counter to 0: toggle on, run once, toggle off.
`debug`	BOOLEAN	false	Log details.

Outputs: INT. Re-executes every queue (necessary for the counter to advance).

Random Select¶

Randomly forwards one of up to 5 connected inputs (any type) to the output.

Only connected inputs are candidates. With seed = -1 the pick is fresh each run; a non-negative seed makes it reproducible.

Parameter	Type	Default	Description
`input_1`–`input_5`	any	(optional)	Candidate values.
`seed`	INT	-1	`-1` = unseeded; `≥ 0` = reproducible selection.

Outputs: output (the chosen value), selected_index (1-based; 0 if nothing connected).

Gate Any¶

Pass any value through when enabled; when disabled, everything downstream of the gate is silently skipped.

Drop it on any connection to toggle a whole branch on and off. With skip_upstream on, closing the gate also prevents the nodes feeding value from executing at all (lazy evaluation), so expensive upstream work is skipped too.

Parameter	Type	Default	Description
`value`	any	—	The value to pass through when the gate is open.
`enabled`	BOOLEAN	true	When off, downstream nodes are silently skipped.
`skip_upstream`	BOOLEAN	true	When the gate is closed, also skip the branch feeding `value`.

Outputs: value (passed through when enabled).

Float Lerp¶

Unclamped linear interpolation: a + (b − a) · t.

t outside 0–1 extrapolates on purpose.

Parameter	Type	Default	Description
`a`	FLOAT	0.0	Value at `t = 0`.
`b`	FLOAT	1.0	Value at `t = 1`.
`t`	FLOAT	0.0	Blend factor, unclamped.

Outputs: FLOAT.

Variables¶

The Property nodes give a graph named variables held in server memory, in two scopes: workflow (temporary, per-run working values) and session (persists until the server restarts or is explicitly cleared).

Property¶

A named variable in one node: wire a value into value to set it (and pass it through), leave value unwired to get it — teleporting values across a graph without wires.

Setter and getter are the same node with the same name. Because execution order between a setter and getter isn't guaranteed by name alone, wire the setter's trigger_out into the getter's trigger when ordering matters. A getter that finds nothing returns None by default (useful on first runs), or raises with a list of available keys when error_if_missing is on. If the key isn't in the chosen scope, the other scope is checked as a fallback.

Parameter	Type	Default	Description
`name`	STRING	`my_var`	The variable key; setter and getter must match.
`session_scope`	BOOLEAN	false	Off = workflow scope; on = session scope.
`error_if_missing`	BOOLEAN	false	Raise instead of returning `None` when the key is absent.
`value`	any	(optional)	Connected = set; disconnected = get.
`trigger`	any	(optional)	Execution-order dependency, passed through to `trigger_out`.
`debug`	BOOLEAN	false	Log store operations.

Outputs: value (the stored or retrieved value), trigger_out.

Property Clear¶

Clears stored properties — a whole scope, or a single key.

Parameter	Type	Default	Description
`scope`	choice	—	`workflow`, `session`, or `all`.
`key`	STRING	`*`	A specific key, or `*` for every key in the scope.
`trigger`	any	(optional)	Execution-order dependency.
`debug`	BOOLEAN	false	Log what was cleared.

Outputs: trigger_out.

Property List¶

Lists all stored properties with short value summaries (tensor shapes, dict keys, truncated strings), per scope.

Parameter	Type	Default	Description
`scope`	choice	—	`workflow`, `session`, or `all`.
`trigger`	any	(optional)	Execution-order dependency.

Outputs: listing (the report), trigger_out.

Timing & logging¶

Timer Start / Timer Stop¶

Measure wall-clock time between two points in a graph.

Timer Start outputs a timestamp; wire it into Timer Stop's start_time, and wire the output of the work you're measuring into Timer Stop's trigger so it evaluates after that work. The elapsed time comes out both human-readable (4m32s) and as raw seconds, and is printed to the console.

Timer Start — no inputs. Outputs: start_time (FLOAT).

Timer Stop:

Parameter	Type	Default	Description
`start_time`	FLOAT (input)	—	Wired from Timer Start.
`trigger`	any	(optional)	Forces evaluation after the measured work; value ignored.

Outputs: formatted (e.g. 1h02m05s), seconds.

Print¶

Prints a custom message to the server console when it executes.

Connect anything to trigger to control when it fires and to pass that value onward.

Parameter	Type	Default	Description
`text`	STRING	`Hello`	The message (printed as `[CCN] <text>`).
`trigger`	any	(optional)	Execution-order dependency, passed through.

Outputs: trigger (passed through).

Inspect Tensor¶

Type-aware inspection of any value — tensors get shape/dtype/device and min/max/mean/std, dicts get per-key summaries — printed to the console and returned as a string, with the input passed through.

Insert it inline anywhere without breaking a connection.

Parameter	Type	Default	Description
`label`	STRING	`Tensor`	Report header.
`data`	any	(optional)	The value to inspect.

Outputs: passthrough (the input, unchanged), info (the report).

Files & tokens¶

Load JSON File¶

Loads a .json file from ComfyUI's input directory (chosen from a dropdown) and returns its raw text plus the filename.

The dropdown lists JSON files found recursively under the input directory. Content is validated as JSON — invalid files return an empty string. Edits to the file on disk re-trigger execution automatically.

Parameter	Type	Default	Description
`json_file`	choice	—	A JSON file from the input directory.

Outputs: json_string, filename.

Load JSON File Path¶

The same JSON loader, but for an arbitrary path typed into a string field.

Parameter	Type	Default	Description
`file_path`	STRING	`""`	Absolute or relative path to a JSON file.

Outputs: json_string, filename.

Token Counter¶

Counts how many tokens a prompt uses, against the connected model's token budget.

With a CLIP connected it uses the real tokenizer and reports used/capacity (percent) with warnings when nearing or hitting the limit (SD-style CLIP has a ~77-token window; T5 encoders like Wan/Flux allow ~512). Without CLIP it falls back to a rough ~4-characters-per-token estimate.

Parameter	Type	Default	Description
`text`	STRING	`""`	The prompt.
`clip`	CLIP	(optional)	Enables an exact count.

Outputs: token_count, info.

Conditioning Token Count¶

Reads the sequence length — the number of token positions — directly from a conditioning tensor.

Parameter	Type	Default	Description
`conditioning`	CONDITIONING	—	The conditioning to measure.

Outputs: sequence_length, info (includes hidden dim and full shape).