# Workflows vs Forge

Layer gives you two ways to generate assets through the API. They share the same [async lifecycle](/docs/async-jobs) and [Creative Unit](/docs/creative-units) model, but suit different jobs.

## Forge — direct model execution

**Forge** runs a single model with full control over its parameters. Reach for it when you know exactly which model you want and want to drive it directly.

* Submit: `POST /v1/workspaces/{id}/inferences` → `inference_id`
* Poll: `GET /v1/workspaces/{id}/inferences/{id}`
* Parameters: prompt, dimensions, batch size, style `weight`, [guidance files](/docs/files) (reference images, masks, poses, depth maps), video/3D/audio options, and post-processing (upscale, vectorize, remove background).

## Workflows — Blueprint pipelines

A **workflow** is a multi-step pipeline built visually in the Layer app’s Blueprint editor — for example: generate an image → upscale → remove background. It’s callable as a single API run, with inputs and outputs defined by the Blueprint.

* Submit: `POST /v1/workspaces/{id}/workflows/{workflow_id}/runs` → `run_id`
* Poll: `GET /v1/workspaces/{id}/workflows/{workflow_id}/runs/{run_id}`
* The run response includes a per-step `steps[]` array so you can see progress and pinpoint a failing step.

## Choosing

| Use **Forge** when…                          | Use a **Workflow** when…                                |
| -------------------------------------------- | ------------------------------------------------------- |
| You want one model with specific parameters. | You need several steps chained together.                |
| Your logic lives in your own code.           | Your team already built the pipeline as a Blueprint.    |
| You want to vary model/params per request.   | You want a stable, named pipeline callable as one unit. |

Both are asynchronous — see [Async jobs & polling](/docs/async-jobs). Note that [rate limits](/docs/rate-limits) apply to forge generation; workflow runs are not currently subject to those per-minute generation limits.