User-Authored Presets

CodeComfy ships with two built-in presets (hq-image and hq-video), but you are not stuck with them. Any ComfyUI workflow JSON you drop into .codecomfy/presets/ in your workspace becomes a first-class preset that appears in the generation commands.

Where presets live

User-authored presets live at .codecomfy/presets/*.json in your workspace root. One file per preset. CodeComfy picks them up automatically on activation — no reload needed after you add the file.

your-workspace/
  .codecomfy/
    presets/
      my-fast-draft.json
      my-painterly-landscape.json

How to create a preset

The quickest path is to start from a working HQ preset and edit from there.

1. Open the Command Palette (Ctrl+Shift+P).
2. Run CodeComfy: Create Preset from HQ Template.
3. Pick image or video. The template preset is written to .codecomfy/presets/ with placeholders you can edit.
4. Adjust the fields (id, name, defaults, workflow). Save the file.

The preset appears in the next generation pick list.

You can also hand-author a preset from scratch. See the shape below.

Preset shape

A preset is a single JSON object with the following fields:

Field	Required	Type	Purpose
id	yes	string	Unique identifier. Letters, digits, _, - only.
name	yes	string	Human-readable name shown in the QuickPick.
description	no	string	Optional one-liner rendered next to the name.
kind	yes	"image" or "video"	Determines which command surfaces the preset.
defaults	yes	object	Default input values (width, height, steps, cfg_scale, fps for video, etc).
workflow	no	object	ComfyUI workflow JSON — a node-id → node object mapping.

defaults are merged with the inputs you provide at generation time. The user-provided prompt always wins over whatever is in defaults.

The workflow field

The workflow object is a raw ComfyUI workflow JSON — the same shape you get from ComfyUI’s Save (API Format) export. Each key is a node id; each value is an object with a class_type and an inputs map.

Reference: ComfyUI workflow JSON docs.

CodeComfy auto-injects values into the following node types so your prompt, seed, and dimensions flow through without you having to hard-code them:

Node class_type	What gets injected
CLIPTextEncode	text (positive + negative prompt, by node wiring)
KSampler	seed, steps, cfg, when supplied
EmptyLatentImage	width, height
CheckpointLoaderSimple	ckpt_name (if codecomfy.defaultCheckpoint is set)

Node types outside this list (e.g. VAELoader, SaveImage, custom nodes) are used as-authored — CodeComfy does not touch them. Put any checkpoint-specific, LoRA, or custom-node configuration into those untouched nodes.

Example preset

A minimal text-to-image preset, 5 nodes:

{
  "id": "my-fast-draft",
  "name": "My Fast Draft",
  "description": "Lower-step image preset for quick iteration.",
  "kind": "image",
  "defaults": {
    "width": 768,
    "height": 768,
    "steps": 15,
    "cfg_scale": 5.5
  },
  "workflow": {
    "1": {
      "class_type": "CheckpointLoaderSimple",
      "inputs": { "ckpt_name": "sd_xl_base_1.0.safetensors" }
    },
    "2": {
      "class_type": "CLIPTextEncode",
      "inputs": { "clip": ["1", 1], "text": "" }
    },
    "3": {
      "class_type": "EmptyLatentImage",
      "inputs": { "width": 768, "height": 768, "batch_size": 1 }
    },
    "4": {
      "class_type": "KSampler",
      "inputs": {
        "model": ["1", 0],
        "positive": ["2", 0],
        "negative": ["2", 0],
        "latent_image": ["3", 0],
        "seed": 0,
        "steps": 15,
        "cfg": 5.5,
        "sampler_name": "dpmpp_2m",
        "scheduler": "karras",
        "denoise": 1.0
      }
    },
    "5": {
      "class_type": "SaveImage",
      "inputs": { "images": ["4", 0], "filename_prefix": "codecomfy" }
    }
  }
}

The CLIPTextEncode text field is empty because CodeComfy fills it with your prompt at generation time.

Validation

CodeComfy registers a JSON schema for .codecomfy/presets/*.json via the jsonValidation contribution. That means VS Code shows errors inline while you are editing:

• Missing required fields (id, name, kind, defaults) are flagged.
• Invalid id characters show a squiggle.
• Wrong kind values ("audio", typos) show a dropdown of valid options.
• Node objects missing a class_type are flagged.

If the file validates in the editor, it will load at activation. If a required field is missing or malformed, the preset is skipped with a WARN in the CodeComfy output channel.

Troubleshooting

Preset does not appear in the QuickPick. Check the CodeComfy output channel (CodeComfy: Open Output Channel) for a preset-skip WARN. Common causes: duplicate id, missing kind, or the JSON itself is invalid.

Generation fails with “node type not supported”. The workflow contains a custom node that your ComfyUI install does not have. Install the custom node in ComfyUI or remove it from the preset.

Checkpoint not found. Either set codecomfy.defaultCheckpoint to a checkpoint you have, or edit the ckpt_name inside the preset’s CheckpointLoaderSimple node directly.