KComfy

Kotlin client for interacting with ComfyUI-style backends — render ComfyUI workflows from Kotlin data classes and manage prompt jobs programmatically.

Table of contents

• Features
• Quickstart
• Installation
• Usage
• Workflows

Features

• Render ComfyUI JSON workflows from Mustache templates and Kotlin data classes
• Retrofit-based HTTP client with WebSocket job updates
• Example workflows and templates included
• Simple job lifecycle management with automatic cleanup

Quickstart

1. Add KComfy to your project (see Installation).
2. Start a ComfyUI-compatible server (e.g., local ComfyUI on port 8000).
3. Use KComfy to render a workflow and submit a prompt.

Installation

Kotlin (Gradle Kotlin DSL)

repositories {
    maven {
        url = uri("https://maven.ivcode.org.s3.amazonaws.com/snapshot")
    }
}

dependencies {
    implementation("org.ivcode:kcomfy:0.1-SNAPSHOT")
}

Usage

Minimal example (quick)

val kComfy = KComfy.create("http://localhost:8000")

kComfy.use { api ->
  val job = api.prompt(CheckpointTextToImage(
    checkpoint = "sdXL_v10VAEFix",
    prompt = "a happy astronaut riding a horse in space, cartoon, vibrant"
  ))

  job.use {
    val outputs = it.getOutputs().get()   // wait and fetch outputs
    outputs.forEach { output ->
      api.view(output).use { input ->
        File(output.filename).outputStream().use { fileOut ->
          input.copyTo(fileOut)
        }
      }
    }
  }
}

Note: This example assumes you have the checkpoint "sdXL_v10VAEFix" available on your ComfyUI server.

Workflows

Workflows are Mustache templates plus Kotlin model classes. Templates should be on the classpath (e.g., src/main/resources) and are rendered using property names from the data class.

Example file layout:

src/
└── main/
    ├── kotlin/
    │   └── workflows/
    │       └── CheckpointTextToImage.kt
    └── resources/
        └── comfy/
            └── checkpoint_text_to_image.json.mustache

Example model and template annotation:

@WorkflowTemplate("comfy/checkpoint_text_to_image.json.mustache")
data class CheckpointTextToImage(
    var checkpoint: String,
    var prompt: String,
    var negativePrompt: String = "",
    var width: Int = 1024,
    var height: Int = 1024,
    var batch: Int = 1,
    var steps: Int = 30,
    var cfg: Double = 7.0,
    var sampler: SamplerName = SamplerName.EULER,
    var scheduler: Scheduler = Scheduler.NORMAL,
    var seed: Long = Random.nextLong(0, Long.MAX_VALUE),
    var denoise: Double = 1.0
)

Template notes:

• Templates are rendered using Mustache; {{prompt}} maps to the prompt property of the data class.
• Place templates in your resources so they are on the runtime classpath.
• See the checkpoint_text_to_image.json.mustache template file to see how the data class properties maps to the template.