each::workflows Overview

What is each::workflows?

A workflow is a saved recipe of AI steps that runs with one API call. You choose the models, prompts and post-processing once. Your app sends the inputs (usually a photo or a short text) and gets the finished result back.

photo ──▶ [ edit image ▶ animate ▶ resize ▶ add music ] ──▶ ready-to-post video
           one workflow, one API call

Why use a workflow?

You want to…	A workflow gives you
Ship an AI feature as one call	Prompts, models and intermediate files stay on Eachlabs. Your app only sends inputs.
Change it without an app release	Edit a prompt or swap a model in a new version; callers keep the same workflow ID.
Keep working when a model fails	A `fallback` model runs automatically if the first one errors.
Combine image, video, audio and text	Chain an image edit into a video model, then resize, merge clips or add music.
Go faster	Run independent steps (one per photo, one per scene) in `parallel`.
Let AI decide the path	Ask a vision model a question, then route with a `choice` step.
Use your own systems	Call any public HTTPS API from a step to fetch data or run your own code.

Workflows are free. You pay only for the model steps that run.

If you call one model with a prompt your user writes, a direct prediction is simpler. Use a workflow once the prompt is fixed, you want a fallback, or there is a second step.

Authoring rules

Follow these and a workflow runs on the first try.

Reference values with $. paths: $.inputs.<name> for inputs, $.<step_id>.primary or $.<step_id>.output for earlier steps, brackets for arrays ($.step.output[0]). {{inputs.prompt}} is not a reference; it is sent to the model as literal text.
A value that is exactly one reference keeps its type (URL, number, object). A reference inside a longer string is inserted as text: "A portrait of $.inputs.name in neon light".
primary is the main result. For image and video models it is the file URL. For eachlabs-llm-router it is the chat completion, so the text is at $.<step_id>.primary.choices[0].message.content. For eachlabs-http-step it is the response, so JSON fields are at $.<step_id>.primary.output.<field>.
Prompt fields quote whole references. If a prompt-like field (prompt, text, content, message, caption, title, description) is exactly $.step.primary or $.step.output, the value arrives wrapped in quotes. Reference a deeper field or embed it in a sentence.
Step types are model, parallel, choice and pass. HTTP requests and Python code are not step types; they run as model steps through eachlabs-http-step and eachlabs-python-step (private beta). See HTTP steps.
Every step needs a unique id, including steps inside branches. A step can only reference inputs and steps that ran before it.
Inputs live in input_schema.properties. Each field takes type (string, number, boolean, image, video, voice for audio, file), required (true/false) and an optional default_value. Files are passed as public URLs. Give every optional input a default_value, because a run fails if a step references an omitted input with no default.
Parallel: each branch is a list of steps. Later steps reference branch steps by their own id.
Choice: a condition, a condition_met_branch and a default_branch, both non-empty. Afterwards, $.<choice_id>.primary is the result of whichever branch ran. Operators: equals, not_equals, greater_than, greater_than_or_equal, less_than, less_than_or_equal, string_equals, string_matches (a * wildcard pattern such as "*.mp4"), exists, is_not_null, combined with and, or and not.
Fallback is set per model step and has its own params, because the second model may take different parameters.
The result of a run is the output of its last step. To return named values, add "metadata": {"output_mapping": {"video": "$.final.primary"}} to the definition.
Models are checked when a step runs, not when you save. Read a model’s inputs with GET /v1/models/{slug} and test the workflow once before shipping it.

Create, run and read

# Create: returns workflow_id; the first version is "v1"
curl -X POST https://api.eachlabs.ai/v1/workflows \
  -H "Authorization: Bearer $EACH_API_KEY" -H "Content-Type: application/json" \
  -d @workflow.json

# Run: returns {"execution_id": "...", "status": "queued"}
curl -X POST https://api.eachlabs.ai/v1/workflows/trigger/$WORKFLOW_ID/v1 \
  -H "Authorization: Bearer $EACH_API_KEY" -H "Content-Type: application/json" \
  -d '{"inputs": {"photo": "https://example.com/selfie.jpg"}}'

# Read: poll until status is completed, failed or cancelled
curl https://api.eachlabs.ai/v1/workflows/executions/$EXECUTION_ID \
  -H "Authorization: Bearer $EACH_API_KEY"

A completed run has output (the last step’s result), workflow_output (your named outputs) and step_outputs.<step_id> for every step. Add "webhook_url" to the run request to be notified instead of polling. If creating returns 400 invalid argument, send the definition to PUT /v1/workflows/{workflowID}/versions/v1 as {"definition": {...}} to see the exact error.

Examples

Each example is a complete workflow.json for the create call. Replace the input URLs with your own files, uploaded with each::storage if they are not public yet.

1. One-photo effect

Turn one photo into a finished clip. The prompt stays on Eachlabs, and a second model takes over if the first one fails.

photo ──▶ moonwalk (reference-to-video, fallback: image-to-video)

{
  "name": "Moonwalk",
  "definition": {
    "input_schema": {
      "properties": {
        "photo": { "type": "image", "required": true }
      }
    },
    "steps": [
      {
        "id": "moonwalk",
        "type": "model",
        "model": "bytedance-seedance-2-0-reference-to-video-fast",
        "params": {
          "prompt": "The person from the reference image, in a white astronaut suit with the helmet visor up, takes slow bouncing steps on the Moon, plants a flag and waves at the camera. Earth rises behind them. Keep their face exactly as in the reference image.",
          "image_urls": ["$.inputs.photo"],
          "duration": "8",
          "resolution": "720p",
          "aspect_ratio": "9:16"
        },
        "fallback": {
          "enabled": true,
          "model": "kling-v3-pro-image-to-video",
          "params": {
            "prompt": "The person is now an astronaut on the Moon, helmet visor up. They wave at the camera as Earth rises behind them.",
            "start_image_url": "$.inputs.photo",
            "duration": "5"
          }
        }
      }
    ]
  }
}

Inputs: {"photo": "https://example.com/selfie.jpg"}

2. Restyle a photo, then animate it

Chain two models: edit the image, then pass the result to a video model.

photo ──▶ restyle (image edit) ──▶ animate (image-to-video)

{
  "name": "Styled Photo to Video",
  "definition": {
    "input_schema": {
      "properties": {
        "photo": { "type": "image", "required": true },
        "style": { "type": "string", "required": false, "default_value": "a 1990s film photo with warm grain" }
      }
    },
    "steps": [
      {
        "id": "restyle",
        "type": "model",
        "model": "nano-banana-2-edit",
        "params": {
          "prompt": "Restyle this photo as $.inputs.style. Keep the person's face, identity and pose unchanged.",
          "image_urls": ["$.inputs.photo"],
          "aspect_ratio": "9:16"
        },
        "fallback": {
          "enabled": true,
          "model": "gpt-image-v2-edit",
          "params": {
            "prompt": "Restyle this photo as $.inputs.style. Keep the person's face, identity and pose unchanged.",
            "image_urls": ["$.inputs.photo"],
            "image_size": "1024x1536"
          }
        }
      },
      {
        "id": "animate",
        "type": "model",
        "model": "kling-v3-pro-image-to-video",
        "params": {
          "start_image_url": "$.restyle.primary",
          "prompt": "The person smiles and turns slightly toward the camera. Gentle handheld motion, natural light.",
          "duration": "5"
        }
      }
    ]
  }
}

Inputs: {"photo": "https://example.com/portrait.jpg", "style": "a watercolor illustration"} $.restyle.primary is the edited image, from the fallback model if it ran. style is optional because it has a default.

3. An LLM writes the prompt

A vision LLM looks at the product and writes the scene. The image and video models follow it.

person + product ──▶ write_scene (LLM) ──▶ photo (image edit) ──▶ video (image-to-video)

{
  "name": "UGC Product Video",
  "definition": {
    "input_schema": {
      "properties": {
        "person": { "type": "image", "required": true },
        "product": { "type": "image", "required": true },
        "setting": { "type": "string", "required": false, "default_value": "a bright modern kitchen" }
      }
    },
    "steps": [
      {
        "id": "write_scene",
        "type": "model",
        "model": "eachlabs-llm-router",
        "params": {
          "model": "google/gemini-3.6-flash",
          "messages": [
            { "role": "system", "content": "You write prompts for photorealistic smartphone photos. Reply with one paragraph and nothing else." },
            {
              "role": "user",
              "content": [
                { "type": "text", "text": "Describe a natural photo of a person holding and using this product in $.inputs.setting. Describe the product by how it looks." },
                { "type": "image_url", "image_url": { "url": "$.inputs.product" } }
              ]
            }
          ]
        }
      },
      {
        "id": "photo",
        "type": "model",
        "model": "nano-banana-2-edit",
        "params": {
          "prompt": "Use the person from the first image and the product from the second image. $.write_scene.primary.choices[0].message.content",
          "image_urls": ["$.inputs.person", "$.inputs.product"],
          "aspect_ratio": "9:16"
        }
      },
      {
        "id": "video",
        "type": "model",
        "model": "kling-v3-pro-image-to-video",
        "params": {
          "start_image_url": "$.photo.primary",
          "prompt": "Handheld smartphone video. The person shows the product to the camera, smiles and uses it naturally.",
          "duration": "5",
          "generate_audio": true
        }
      }
    ]
  }
}

Inputs: {"person": "https://example.com/creator.jpg", "product": "https://example.com/honey.png"}

4. Two photos, one clip, with music

Prepare both photos at the same time, put both people in one video, then resize it and add music.

             ┌─ edit_a ─┐
photos ──▶ prepare      ├──▶ clip (reference-to-video) ──▶ resize ──▶ soundtrack
             └─ edit_b ─┘

{
  "name": "Travel Buddies Clip",
  "definition": {
    "input_schema": {
      "properties": {
        "photo_a": { "type": "image", "required": true },
        "photo_b": { "type": "image", "required": true },
        "music": { "type": "voice", "required": true }
      }
    },
    "steps": [
      {
        "id": "prepare",
        "type": "parallel",
        "branches": [
          {
            "steps": [
              {
                "id": "edit_a",
                "type": "model",
                "model": "nano-banana-2-edit",
                "params": {
                  "prompt": "Full-body photo of this person as a traveler with a backpack in a sunny old-town street. Keep their face and identity.",
                  "image_urls": ["$.inputs.photo_a"],
                  "aspect_ratio": "9:16"
                }
              }
            ]
          },
          {
            "steps": [
              {
                "id": "edit_b",
                "type": "model",
                "model": "nano-banana-2-edit",
                "params": {
                  "prompt": "Full-body photo of this person as a traveler with a camera in a sunny old-town street. Keep their face and identity.",
                  "image_urls": ["$.inputs.photo_b"],
                  "aspect_ratio": "9:16"
                }
              }
            ]
          }
        ]
      },
      {
        "id": "clip",
        "type": "model",
        "model": "bytedance-seedance-2-0-reference-to-video-fast",
        "params": {
          "prompt": "The two people from the reference images walk side by side through a sunny old-town market, laughing and taking a selfie together.",
          "image_urls": ["$.edit_a.primary", "$.edit_b.primary"],
          "duration": "8",
          "resolution": "720p",
          "aspect_ratio": "9:16",
          "generate_audio": false
        }
      },
      {
        "id": "resize",
        "type": "model",
        "model": "scale-video",
        "params": { "video_url": "$.clip.primary", "width": 1080, "height": 1920, "mode": "crop" }
      },
      {
        "id": "soundtrack",
        "type": "model",
        "model": "ffmpeg-api-merge-audio-video",
        "params": { "video_url": "$.resize.primary", "audio_url": "$.inputs.music" }
      }
    ]
  }
}

Inputs:

{"photo_a": "https://example.com/a.jpg", "photo_b": "https://example.com/b.jpg", "music": "https://example.com/track.mp3"}

Both edits run at the same time. To use the same music every time, replace $.inputs.music with a fixed URL.

5. A reliable vision call

Write alt text for a product photo. If the first provider fails, a second one answers.

image ──▶ describe (LLM, fallback: another provider) ──▶ alt_text

{
  "name": "Product Alt Text",
  "definition": {
    "input_schema": {
      "properties": {
        "image": { "type": "image", "required": true }
      }
    },
    "steps": [
      {
        "id": "describe",
        "type": "model",
        "model": "eachlabs-llm-router",
        "params": {
          "model": "google/gemini-3.6-flash",
          "temperature": 0,
          "messages": [
            {
              "role": "user",
              "content": [
                { "type": "text", "text": "Write alt text for this product photo in one sentence of at most 20 words." },
                { "type": "image_url", "image_url": { "url": "$.inputs.image" } }
              ]
            }
          ]
        },
        "fallback": {
          "enabled": true,
          "model": "eachlabs-llm-router",
          "params": {
            "model": "openai/gpt-4.1-mini",
            "temperature": 0,
            "messages": [
              {
                "role": "user",
                "content": [
                  { "type": "text", "text": "Write alt text for this product photo in one sentence of at most 20 words." },
                  { "type": "image_url", "image_url": { "url": "$.inputs.image" } }
                ]
              }
            ]
          }
        }
      }
    ],
    "metadata": {
      "output_mapping": { "alt_text": "$.describe.primary.choices[0].message.content" }
    }
  }
}

Inputs: {"image": "https://cdn.dummyjson.com/product-images/groceries/honey-jar/1.webp"} The run’s workflow_output.alt_text holds the sentence, whichever provider answered.

6. Call an API, then use its data

Fetch a product from an API and turn it into an ad image. This example calls the public DummyJSON test API; point url at your own API.

product_id ──▶ product (HTTP GET) ──▶ ad (image edit)

{
  "name": "Product Ad From API",
  "definition": {
    "input_schema": {
      "properties": {
        "product_id": { "type": "string", "required": true }
      }
    },
    "steps": [
      {
        "id": "product",
        "type": "model",
        "model": "eachlabs-http-step",
        "params": {
          "url": "https://dummyjson.com/products/$.inputs.product_id",
          "method": "GET",
          "timeout_seconds": 20,
          "fail_on_status": "non_2xx"
        }
      },
      {
        "id": "ad",
        "type": "model",
        "model": "nano-banana-2-edit",
        "params": {
          "prompt": "Studio advertising photo of this $.product.primary.output.title on a clean pastel background with soft shadows. Leave empty space at the top for a headline.",
          "image_urls": ["$.product.primary.output.images[0]"],
          "aspect_ratio": "4:5"
        }
      }
    ]
  }
}

Inputs: {"product_id": "27"}

eachlabs-http-step takes url, method, headers, query_params, body (a JSON object for POST, PUT or PATCH), timeout_seconds (up to 30 recommended) and fail_on_status.
Set fail_on_status to "non_2xx" so an error response stops the run.
To run your own code, deploy it as an HTTPS endpoint (a serverless function is enough), POST the inputs to it, and read its JSON reply in the next step. Organizations in the private beta can use the eachlabs-python-step model instead.
Only public URLs are reachable. Definitions are stored, so use scoped, read-only tokens in headers.

7. Let AI choose the path

Ask a vision model what the photo shows, then pick the matching recipe. The next step continues from whichever branch ran.

photo ──▶ subject (LLM) ──▶ portrait (choice) ─┬─ pet_portrait ────┐
                                               └─ person_portrait ─┴──▶ animate

{
  "name": "Pet or Person Portrait",
  "definition": {
    "input_schema": {
      "properties": {
        "photo": { "type": "image", "required": true }
      }
    },
    "steps": [
      {
        "id": "subject",
        "type": "model",
        "model": "eachlabs-llm-router",
        "params": {
          "model": "google/gemini-3.6-flash",
          "temperature": 0,
          "messages": [
            {
              "role": "user",
              "content": [
                { "type": "text", "text": "Is the main subject of this photo an animal or a person? Answer with one word in capitals: PET or PERSON." },
                { "type": "image_url", "image_url": { "url": "$.inputs.photo" } }
              ]
            }
          ]
        }
      },
      {
        "id": "portrait",
        "type": "choice",
        "condition": {
          "expression": "$.subject.primary.choices[0].message.content",
          "operator": "string_matches",
          "value": "*PET*"
        },
        "condition_met_branch": {
          "name": "pet",
          "steps": [
            {
              "id": "pet_portrait",
              "type": "model",
              "model": "nano-banana-2-edit",
              "params": {
                "prompt": "A royal oil-painting portrait of this animal wearing a small crown and a velvet cape, in a gilded frame.",
                "image_urls": ["$.inputs.photo"],
                "aspect_ratio": "4:5"
              }
            }
          ]
        },
        "default_branch": {
          "name": "person",
          "steps": [
            {
              "id": "person_portrait",
              "type": "model",
              "model": "nano-banana-2-edit",
              "params": {
                "prompt": "A royal oil-painting portrait of this person wearing a crown and a velvet cape, in a gilded frame. Keep their face unchanged.",
                "image_urls": ["$.inputs.photo"],
                "aspect_ratio": "4:5"
              }
            }
          ]
        }
      },
      {
        "id": "animate",
        "type": "model",
        "model": "kling-v3-pro-image-to-video",
        "params": {
          "start_image_url": "$.portrait.primary",
          "prompt": "The painting comes to life: the subject blinks, looks around and the cape moves slightly.",
          "duration": "5"
        }
      }
    ]
  }
}

Inputs: {"photo": "https://example.com/dog.jpg"} string_matches uses * wildcards, so *PET* also matches PET.. Use and, or and not to combine conditions.

8. Several scenes in parallel, joined with music

Each branch builds one scene (image, then video). All scenes render at the same time, then the clips are joined and music is added.

                     ┌─ scene1_image ─▶ scene1_video ─┐
product ──▶ scenes ──┼─ scene2_image ─▶ scene2_video ─┼──▶ merge ──▶ soundtrack
                     └─ scene3_image ─▶ scene3_video ─┘

{
  "name": "Three-Scene Product Story",
  "definition": {
    "input_schema": {
      "properties": {
        "product": { "type": "image", "required": true },
        "product_name": { "type": "string", "required": true },
        "music": { "type": "voice", "required": true }
      }
    },
    "steps": [
      {
        "id": "scenes",
        "type": "parallel",
        "branches": [
          {
            "steps": [
              {
                "id": "scene1_image",
                "type": "model",
                "model": "nano-banana-2-edit",
                "params": {
                  "prompt": "Morning: the $.inputs.product_name on a sunny breakfast table next to fresh bread. Keep the product exactly as in the photo.",
                  "image_urls": ["$.inputs.product"],
                  "aspect_ratio": "9:16"
                }
              },
              {
                "id": "scene1_video",
                "type": "model",
                "model": "kling-v3-standard-image-to-video",
                "params": { "start_image_url": "$.scene1_image.primary", "prompt": "Slow push-in on the product.", "duration": "5" }
              }
            ]
          },
          {
            "steps": [
              {
                "id": "scene2_image",
                "type": "model",
                "model": "nano-banana-2-edit",
                "params": {
                  "prompt": "Afternoon: the $.inputs.product_name on a picnic blanket in a park. Keep the product exactly as in the photo.",
                  "image_urls": ["$.inputs.product"],
                  "aspect_ratio": "9:16"
                }
              },
              {
                "id": "scene2_video",
                "type": "model",
                "model": "kling-v3-standard-image-to-video",
                "params": { "start_image_url": "$.scene2_image.primary", "prompt": "Leaves move in the wind, slow orbit around the product.", "duration": "5" }
              }
            ]
          },
          {
            "steps": [
              {
                "id": "scene3_image",
                "type": "model",
                "model": "nano-banana-2-edit",
                "params": {
                  "prompt": "Evening: the $.inputs.product_name on a kitchen counter lit by a warm lamp. Keep the product exactly as in the photo.",
                  "image_urls": ["$.inputs.product"],
                  "aspect_ratio": "9:16"
                }
              },
              {
                "id": "scene3_video",
                "type": "model",
                "model": "kling-v3-standard-image-to-video",
                "params": { "start_image_url": "$.scene3_image.primary", "prompt": "Slow pull-back as the lamp glows.", "duration": "5" }
              }
            ]
          }
        ]
      },
      {
        "id": "merge",
        "type": "model",
        "model": "merge-videos",
        "params": {
          "video_urls": ["$.scene1_video.primary", "$.scene2_video.primary", "$.scene3_video.primary"]
        }
      },
      {
        "id": "soundtrack",
        "type": "model",
        "model": "ffmpeg-api-merge-audio-video",
        "params": { "video_url": "$.merge.primary", "audio_url": "$.inputs.music" }
      }
    ]
  }
}

Inputs:

{"product": "https://cdn.dummyjson.com/product-images/groceries/honey-jar/1.webp", "product_name": "honey jar", "music": "https://example.com/track.mp3"}

Steps inside a branch run in order; the branches run at the same time, so three scenes take about as long as one. merge-videos joins the clips in the listed order.

Checklist for agents

Every step has a unique id; every $. reference points to an input or an earlier step.
References use $., never {{...}}.
Every optional input has a default_value.
Model slugs and parameters come from GET /v1/models/{slug}.
LLM text is read from primary.choices[0].message.content; HTTP JSON from primary.output.<field>.
The last step produces the final result, or metadata.output_mapping names it.
You ran it once and checked step_outputs before giving it to users.

Endpoints

All paths are relative to https://api.eachlabs.ai.

Method	Path	Description
`POST`	`/v1/workflows`	Create a workflow
`GET`	`/v1/workflows/{workflowID}`	Get workflow details
`PUT`	`/v1/workflows/{workflowID}`	Rename a workflow
`PUT`	`/v1/workflows/{workflowID}/versions/{versionID}`	Create/update a version
`POST`	`/v1/workflows/trigger/{workflowID}/{versionID}`	Trigger execution
`POST`	`/v1/workflows/bulk-trigger/{workflowID}/{versionID}`	Bulk trigger (up to 10)
`GET`	`/v1/workflows/{workflowID}/executions`	List executions
`GET`	`/v1/workflows/executions/{executionID}`	Get execution details
`GET`	`/v1/public/@{nickname}/workflows/{slug}/versions/{versionID}`	Get public workflow (no auth)
`POST`	`/v1/public/@{nickname}/workflows/{slug}/versions/{versionID}/trigger`	Trigger public workflow (`api_key` in JSON body)

Pricing

Workflows are free to create, manage and run. You pay only for the models inside your steps, billed through each::api.

Overview

Concepts

API Reference

OpenAPI Reference

Webhooks

Examples

each::workflows Overview

What is each::workflows?

Why use a workflow?

Authoring rules

Create, run and read

Examples

1. One-photo effect

2. Restyle a photo, then animate it

3. An LLM writes the prompt

4. Two photos, one clip, with music

5. A reliable vision call

6. Call an API, then use its data

7. Let AI choose the path

8. Several scenes in parallel, joined with music

Checklist for agents

Endpoints

Pricing

​What is each::workflows?

​Why use a workflow?

​Authoring rules

​Create, run and read

​Examples

​1. One-photo effect

​2. Restyle a photo, then animate it

​3. An LLM writes the prompt

​4. Two photos, one clip, with music

​5. A reliable vision call

​6. Call an API, then use its data

​7. Let AI choose the path

​8. Several scenes in parallel, joined with music

​Checklist for agents

​Endpoints

​Pricing

What is each::workflows?

Why use a workflow?

Authoring rules

Create, run and read

Examples

1. One-photo effect

2. Restyle a photo, then animate it

3. An LLM writes the prompt

4. Two photos, one clip, with music

5. A reliable vision call

6. Call an API, then use its data

7. Let AI choose the path

8. Several scenes in parallel, joined with music

Checklist for agents

Endpoints

Pricing