Quickstart

Three steps of setup, then one request. Nothing here needs a sales call or a subscription.

Set up

Create an account and add credit on the credit page. Credit is spent per request.
Create an API key under API keys. The secret is shown once and never again.
Pick a model id from the live catalog, which shows prices and how many providers are online right now.

Make a request

POST /v1/workflowsWORKFLOWS · mayhem_comfy_workflows +

Working request

curl https://api.openmayhem.ai/v1/workflows \
  -H "Authorization: Bearer $OPENMAYHEM_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: $(uuidgen)" \
  -d '{
  "model": "image.edit.le2mp",
  "workflow": {},
  "input_files": [
    {
      "filename": "refs/input.png",
      "kind": "image",
      "content_type": "image/png",
      "encoding": "base64",
      "data": "<BASE64_IMAGE>"
    }
  ],
  "response_format": "artifact",
  "timeout_ms": 120000
}'

Parameters

Parameter	Required	Type	Default	Allowed values	Limits and shape
`input_files`	Required	`array`	—	—	min items 1; max items 2 [{"filename":"refs/input.png","kind":"image","content_type":"image/png","encoding":"base64","data":"<BASE64_IMAGE>"}]
`model`	Required	`string`	—	—	—
`workflow`	Required	`object`	—	—	{"node-id":{"class_type":"AllowedNode","inputs":{}}}
`outcome_class`	Optional	`string`	—	—	min length 1; max length 128
`response_format`	Optional	`string`	artifact	artifact, json	{"type":"json_object"}
`runtime_id`	Optional	`string`	—	—	min length 1; max length 128
`timeout_ms`	Optional	`integer`	120000	—	min 1000; max 3600000
`user`	Optional	`string`	—	—	min length 1; max length 512

Input shapes

input_files

{
  "model": "image.edit.le2mp",
  "input_files": [
  {
    "filename": "refs/input.png",
    "kind": "image",
    "content_type": "image/png",
    "encoding": "base64",
    "data": "<BASE64_IMAGE>"
  }
]
}

Jobs are asynchronous. The submit call returns a job. Poll GET /v1/jobs/{job_id}; when it succeeds, download each artifact from its artifact URL.

Workflow surface

Use the complete starter. Node roles, classes, inputs, presets and add-ons below are the values admitted for this workflow.

Workflow nodes

Node ID / role	Class	Inputs
`clip`	`CLIPLoader`	clip_name: part; required; part qwen3vl_4b_fp8_scaled.safetensors device: string; required; fixed default type: string; required; fixed krea2
`decode`	`VAEDecode`	samples: link; required; from sampler[0] vae: link; required; from vae[0]
`denoiser`	`UNETLoader`	unet_name: part; required; part krea2_turbo_fp8_scaled.safetensors weight_dtype: string; required; fixed default
`edit_patch`	`Krea2EditModelPatch`	fit_mode: string; required; values fit model: link; required; from system_lora[0] ref_boost: number; required; min 1; max 4 ref_boost_a: number; required; min 1; max 4 source_image: link; required; from reference_a[0] or scene_scale[0] source_image_b: link; optional; from reference_b[0] source_latent: link; required; from encode_a[0] source_latent_b: link; optional; from encode_b[0] target_latent: link; required; from latent[0] vae: link; required; from vae[0]
`encode_a`	`VAEEncode`	pixels: link; required; from reference_a[0] or scene_scale[0] vae: link; required; from vae[0]
`encode_b`	`VAEEncode`	pixels: link; required; from reference_b[0] vae: link; required; from vae[0]
`latent`	`EmptySD3LatentImage`	batch_size: integer; required; min 1; max 1 height: integer; required; min 256; max 1920; multiple of 64 width: integer; required; min 256; max 1920; multiple of 64
`negative`	`Krea2EditGroundedEncode`	clip: link; required; from clip[0] grounding_px: integer; required; values 768 image: link; required; from reference_a[0] or scene_scale[0] image_b: link; optional; from reference_b[0] prompt: string; required system_prompt: string; required; fixed
`output`	`SaveImage`	filename_prefix: string; required images: link; required; from decode[0]
`positive`	`Krea2EditGroundedEncode`	clip: link; required; from clip[0] grounding_px: integer; required; values 768 image: link; required; from reference_a[0] or scene_scale[0] image_b: link; optional; from reference_b[0] prompt: string; required system_prompt: string; required; fixed
`reference_a`	`LoadImage`	image: string; required
`reference_b`	`LoadImage`	image: string; required
`sampler`	`KSampler`	cfg: number; required; min 1; max 4 denoise: number; required; fixed 1 latent_image: link; required; from latent[0] model: link; required; from edit_patch[0] negative: link; required; from negative[0] positive: link; required; from positive[0] sampler_name: string; required; values euler scheduler: string; required; values simple seed: integer; required; min 0; max 9007199254740991 steps: integer; required; min 1; max 10
`scene_scale`	`ImageScale`	crop: string; required; fixed disabled height: integer; required; fixed 1024 image: link; required; from reference_a[0] upscale_method: string; required; fixed lanczos width: integer; required; fixed 1024
`system_lora`	`LoraLoaderModelOnly`	lora_name: part; required; part Krea2/krea2_identity_edit_v1_2.safetensors model: link; required; from denoiser[0] strength_model: number; required; fixed 1
`vae`	`VAELoader`	vae_name: part; required; part qwen_image_vae.safetensors

Presets

Preset	Applied values
`turbo` Turbo	sampler.steps=10; sampler.cfg=1; sampler.sampler_name=euler; sampler.scheduler=simple; edit_patch.ref_boost=4; edit_patch.ref_boost_a=1; positive.grounding_px=768

Agent drop-ins

This is a media endpoint, so it cannot replace the text model that drives a coding agent. Call the request above from the agent’s HTTP or tool layer.

Routing and trust: Catalog filters, provider constraints, attestation and retail cost limits.