Select Input Type¶

The select input type renders a dropdown menu with a fixed set of predefined options. Each option has a human-readable label (what the user sees) and a machine value (what gets sent to the backend). It is used for parameters with a discrete set of valid choices -- samplers, schedulers, durations, presets, and similar enumerations.

YAML Example¶

Sampler selection from the t2i-dynamic manifest:

inputs:
  - id: sampler_name
    name: "Sampler"
    type: select
    options:
      - {label: "Euler", value: "euler"}
      - {label: "Euler Ancestral", value: "euler_ancestral"}
      - {label: "DPM++ 2M", value: "dpmpp_2m"}
      - {label: "DPM++ SDE", value: "dpmpp_sde"}
      - {label: "DPM++ 2M SDE", value: "dpmpp_2m_sde"}
      - {label: "UniPC", value: "uni_pc"}
      - {label: "LMS", value: "lms"}
    default: "euler"
    tooltip: "Denoising algorithm. Euler is clean and fast. Euler Ancestral adds variety."

Duration selection from the wan22-i2v-fp8 manifest (values are integers, not strings):

  - id: duration
    name: "Duration"
    type: select
    node_id: "4"
    field: "value"
    options:
      - label: "5 seconds"
        value: 81
      - label: "10 seconds"
        value: 161
      - label: "15 seconds"
        value: 241
      - label: "20 seconds"
        value: 321
    default: 81
    tooltip: "Video length before frame interpolation."

Preset selection from the faceid-batch manifest:

  - id: faceid_preset
    name: "FaceID Preset"
    type: select
    options:
      - {label: "FACEID PLUS V2 (recommended)", value: "FACEID PLUS V2"}
      - {label: "FACEID", value: "FACEID"}
      - {label: "FACEID PLUS - SD1.5 only", value: "FACEID PLUS - SD1.5 only"}
      - {label: "FACEID PORTRAIT (style transfer)", value: "FACEID PORTRAIT (style transfer)"}
      - {label: "FACEID PORTRAIT UNNORM - SDXL only (strong)", value: "FACEID PORTRAIT UNNORM - SDXL only (strong)"}
    default: "FACEID PLUS V2"
    tooltip: "Which FaceID model to use."

Fields¶

Field	Type	Required	Default	Description
`id`	string	yes	--	Unique identifier used as the parameter key in form submission and workflow injection.
`name`	string	yes	--	Human-readable label displayed above the dropdown.
`type`	string	yes	--	Must be `select` (or the aliases `dropdown`, `enum`).
`options`	list	yes	--	Array of option objects. Each object has a `label` (displayed text) and a `value` (submitted value). See details below.
`default`	string or number	no	first option	The `value` of the option that should be pre-selected. Must match one of the option values exactly (using string comparison).
`required`	boolean	no	`false`	Whether a selection must be made. Since the first option is always selected by default, this is rarely needed.
`tooltip`	string	no	--	Help text shown on hover over the `?` icon.
`node_id`	string	no	--	(Static workflows only) The ComfyUI node ID where the value is injected.
`field`	string	no	--	(Static workflows only) The input field name on the target node.

Option object format¶

Each item in the options array should be an object with:

Field	Type	Description
`label`	string	What the user sees in the dropdown menu. Can include descriptive text, compatibility notes, or recommendations.
`value`	string or number	What gets sent to the backend. This is the actual value used in the workflow -- it must match what ComfyUI expects.

The label and value can differ significantly. For example, the duration options display "5 seconds" but submit 81 (the frame count at 16fps + 1). This separation lets the UI be human-friendly while the backend receives machine-ready values.

If an option is a plain string instead of an object, both label and value are set to that string.

Frontend Behavior¶

When the runner page encounters a type: select input, the buildSelect() function creates an HTML <select> element:

The element uses the CSS class form-input for consistent styling.
The native browser dropdown arrow is hidden via appearance: none, replaced by a custom SVG chevron positioned on the right side of the element.
Each option from the manifest is rendered as an <option> element, with value set to the option's value and textContent set to the option's label.
The default option is pre-selected by matching String(option.value) === String(input.default). Note the string comparison -- this means default: 81 will match an option with value: 81 even though one is in YAML and the other in the DOM (which stores everything as strings).
Option styling: options have the surface background and text color to match the dark theme.

The data-input-id attribute is set on the select element for form gathering. At submission time, the frontend reads el.value, which returns the value of the currently selected option as a string.

Backend Behavior¶

Static workflows¶

For static workflows, the selected value is injected into the specified node's input field:

node_id: "4"
field: "value"

The backend writes the value directly into workflow["4"]["inputs"]["value"]. For numeric values like frame counts, the form submission and JSON parsing handle the type conversion -- the value arrives as the appropriate type.

Dynamic workflows¶

For dynamic workflows, the selected value is passed as a template variable:

sampler_name = params.get("sampler_name", "euler")
scheduler = params.get("scheduler", "normal")

The value is substituted into block templates:

{
  "sampler": {
    "class_type": "KSampler",
    "inputs": {
      "sampler_name": "{{sampler_name}}",
      "scheduler": "{{scheduler}}"
    }
  }
}

Common select parameters¶

Samplers (denoising algorithms): | Value | Description | |-------|-------------| | euler | Deterministic, clean, fast. Good default. | | euler_ancestral | Adds randomness per step. Good for variety in batches. | | dpmpp_2m | Fast convergence, deterministic. | | dpmpp_sde | Middle ground between deterministic and stochastic. | | dpmpp_2m_sde | SDE variant of DPM++ 2M. | | uni_pc | Predictor-corrector. Good quality at low step counts. | | lms | Linear multi-step. Older algorithm, still viable. |

Schedulers (noise schedules): | Value | Description | |-------|-------------| | normal | Linear noise schedule. Standard default. | | karras | Front-loads denoising. Sharper details. Popular choice. | | exponential | Aggressive early, gentle late. | | sgm_uniform | Uniform spacing. Used by some newer models. | | simple | Simplified schedule. |

Durations (WAN video workflows): | Label | Value (frames) | Calculation | |-------|---------------|-------------| | 5 seconds | 81 | 5 * 16 + 1 | | 10 seconds | 161 | 10 * 16 + 1 | | 15 seconds | 241 | 15 * 16 + 1 | | 20 seconds | 321 | 20 * 16 + 1 |

The +1 comes from WAN's frame encoding convention where the frame count includes the start frame.

Notes¶

The select type is used for both string values (sampler names, preset names) and numeric values (frame counts). The YAML value field determines the type; HTML always stores it as a string, but JSON parsing on the backend restores the original type.
Labels can be much more descriptive than values. Use labels to communicate compatibility constraints (e.g. "LIGHT - SD1.5 only") or recommendations (e.g. "FACEID PLUS V2 (recommended)").
The select type has no search or filtering capability. For large option sets, consider whether the user experience would be better served by a different approach.
The select type also appears nested inside scene_list entries (as per-scene duration dropdowns) and is reused internally to render boolean inputs.
Unlike checkpoint_picker or vae_picker, the select type has a static option list defined in the manifest. It does not query the backend for available choices. Use the picker types when options depend on what is installed on disk.