Start a project from a template

A template is a project configuration someone has already built and tuned: the labeling interface, the quality criteria, the contributor onboarding, and the audience and pricing defaults. You pick one, answer a few questions about your own data, and deploy. Use this when a template matches the job you have; if none does, have the agent build the project from your description instead.

Before you start

You need:

• An account you can sign in to.
• A team with a funded balance. A template run pays real contributors, so the deploy is blocked on an empty balance. See Manage your team and balance.
• Your data, if you have it. You can also let the assistant build the dataset for you, which is what the collection templates do — they generate one empty slot per recording.

Step 1 — Say what you need done

Sign in. The home screen asks What do you need done? Click Data collection if you want data that does not exist yet — video, audio, images, in-the-field capture. Click Data labeling if you already have data and need it annotated, ranked, transcribed, or moderated.

Your choice creates a draft project, opens the Let’s set your task screen, and narrows the templates below to the ones that fit it.

Note

The screen also offers Prompt gisting and Fine tuning. Those are model-side services, not crowd projects, and they lead elsewhere.

Step 2 — Open a template

Under the description field is a Use a template section. Each card names a template and says what it collects. Click the one that matches your job.

Note

The + Project button on the Projects page also creates a draft, but it skips the use-case question, so the section lists every template instead of the ones for your job.

Note

To reuse a pipeline you built in another project, click Import pipeline in the Import section below the templates, and choose the .pipeline file you downloaded with Export. The project opens with that pipeline, but without its dataset or item filter. Attach a dataset, set the filter again if you need one, and check the item count on the Start node before you deploy.

Step 3 — Check the preview, then continue

The card opens a preview: the pipeline the template will build, and a summary of the prompt it will send to the assistant on your behalf — including the defaults it assumes for crowd, price per item, and item length. Read the summary; those defaults are what you will be changing later if they do not fit.

Click Continue. Three things happen at once: the project takes the template’s name, a first pipeline appears already configured from the template, and the assistant starts work on it.

The chat opens with a Started from preset line naming the template. The briefing the template hands the assistant is not shown — the assistant’s own first message covers what you need to know: what the project collects, what the template has already decided for you, and what it needs from you.

Note

The pipeline is built at this point, but it is not yet pointed at any data. That is what the next two steps are for.

Step 4 — Answer the assistant’s questions

The assistant asks everything it needs in one go, right after its opening message. You answer the questions one at a time; some offer suggested answers, and you can type your own instead.

What each template asks:

• Egocentric Activity Video Collection — three questions. How many videos you want (50, 100, 250, 500, or 1000); whether to keep the default list of 13 household activity categories, narrow it, or collect a different domain; and which of the defaults you want changed — audience languages and countries, price per task, time per task, the per-expert cap, and the minimum clip length. “Nothing” is a valid answer to the last one.
• App Crowdtesting — four questions. How many recordings you want (50, 100, 250, or 500, one per tester); the install link testers download the build from; what the app is and whether testers should be on Android or iOS; and what a good test session looks like — plain prose describing what a tester should have done by the time they stop recording, including anything off-limits such as paying or deleting an account.

Answer all of them. The assistant does not touch the pipeline until it has your answers, so this is the cheapest point to change your mind about scope, price, or audience.

Once it starts building, it works through a fixed sequence — create the dataset, point the pipeline at it, apply your answers, validate — and then plays the result back for you to correct or approve in one round.

Step 5 — Attach your data

Upload your dataset, or ask the assistant in the chat to generate one. Templates that collect fresh data — recordings, photos, test sessions — do not need a dataset from you: the assistant creates one row per item you want collected and wires it to the pipeline.

For uploading your own file and setting up its fields, see Create and manage datasets.

Step 6 — Review each node

Open the pipeline and check every node before you launch. The template’s defaults are a starting point, not a recommendation for your budget: price per task, time per task, the audience filters, the labeling interface, and the contributor guidelines are all yours to change.

The expert count and recommended rate shown next to your audience filters tell you whether the audience is large enough and whether your price is in the range that pool expects.

Step 7 — Deploy

Click Deploy.

If anything is wrong, an Issues panel opens instead of launching. Errors — in red — block the deploy. Warnings, in orange, do not.

Click Fix all with agent and the assistant works through the list. Then go back to Step 6, check what it changed, and deploy again.

Once no errors remain, you can launch with the warnings still open: tick I want to deploy with unresolved warnings, then click Deploy.

Caution

A warning that a check has not been run means nobody has verified that part of the configuration. Deploying past it is allowed, but the configuration cannot be changed after launch — so a broken labeling interface or a wrong quality criterion becomes a re-deploy, and the tasks contributors already took are paid for.

Step 8 — Confirm the cost

The cost breakdown opens, line by line, with an expected and a maximum total. If it is more than you want to spend, cancel, go back to Step 6, and lower the price per task or the item count.

Tick the review checkbox and click Confirm and deploy. Billing is pay-as-you-go: each step is charged as it runs, so the final cost can differ from the forecast. See Pricing and billing for what drives each line.

Step 9 — Wait for the launch

Launching Pipeline appears and tracks the deploy step by step. It usually takes 5–15 minutes, and you can close the window and come back.

When it finishes, the tasks are live for contributors.

What’s next

• Export your labeled results when the run has results.
• Resume or cancel a run if you need to stop it.
• Troubleshoot a run if items are not getting done.