Presets and workflows

Ready-made recipes: presets for one file, workflow templates for a whole ad.

Read as Markdown

A preset turns one file into one short video. A workflow template chains several steps into a whole ad.

Presets

Pick a template and add one file. The preset writes the prompt, the scene and the motion, so there is no prompt to write.

PresetWhat you getYou bringLengthSoundTakes per run
Show Your AppYour app screenshot on a real person's phone, animated. VerticalA template, one app screenshotAlways 5 sNone1 to 4
Fashion Try OnYour clothes on a template model, animated. VerticalA template, one photo of the clothes5 or 10 s (5 by default)Optional1 to 4

Each take comes back as its own video, grouped with the others: Variants.

Run a preset

  1. List the templates (which tool or route) and keep the template_id (captpl_) the person picks. Read it fresh each time.
  2. Upload the file and keep its ast_ id: Bring your own files.
  3. Estimate, then submit with config.template_id and the file in reference_images: Price first.
Estimate body: Fashion Try On, 10 seconds
{
  "capability_id": "preset_fashion_tryon",
  "config": {
    "template_id": "captpl_0193c8f0a1b24e7f9d3c5a6b7e8f00b2",
    "reference_images": [
      {
        "assetId": "ast_0193c8f0a1b24e7f9d3c5a6b7e8f0012",
        "alias": "image1",
        "role": "reference"
      }
    ],
    "duration": 10,
    "generate_audio": false
  }
}
  • Fashion Try On takes duration (5 or 10) and an optional generate_audio.
  • Show Your App takes the same body without those two fields.

Preset fields and refusals: Capabilities API.

Workflow templates

A workflow template is a ready-made chain of steps: writers, images, video and captions. One call runs them all. Your plan decides which templates you can run.

TemplateWhat you getYou bring
Creator to cameraAn invented creator talks to camera about your brand, captioned, with music, in 9:16 and 4:5Brand text
Four openings, one adFour finished ads: one body behind four different openings, for A/B testsBrand text
Product photo to spoken adA creator holds your product and talks, then a close-up, with music and captionsBrand text, product photo
Two people: podcast, interview or skitTwo invented people in one scripted scene, captioned, with musicBrand text
Long-form sales videoAbout a minute of one creator selling your offer, with a cutaway, captions and musicOffer text, a voice (the same one on each take)
Studio commercial from one photoTwo short commercials with sound and music, in 9:16 and 16:9Brand text, product photo
Show it working: unboxing, how-to, before and afterA hands-only, narrated product demo with sound effects, music and captionsBrand text, product photo, a voice
Make a finished ad moveYour static ad animated in, held and out, looping, with music, in three shapesOne finished ad image
Sale ad in every sizeA moving offer ad in launch, midway and last-day versions, in four shapesSale text
Your product on the streetThree fake out-of-home stunts with your product, plus one cutBrand and city text, product photo
Try-on with a talking hookA creator hook, then your garment worn by a model, with music and a price cardBrand text, garment photo
Four hooks, four creativesFour statics, each a different hook, in 4:5 and 9:16Brand text, product photo
Rebuild an ad around your productA working ad's layout rebuilt with your brand, two versions in 9:16 and 4:5The ad to remake, product photo, brand text
Store-page photo packSix store-page shots of your product, plus a 4:5 cropProduct text, product photo
One creative, every placementYour ad rebuilt (not cropped) for 1:1, 4:5, 9:16 and 16:9Your ad
App review from a creatorA creator talks up your app or game while your real screen plays on phonesApp text, app screenshot, a voice
Software explainer with a presenterA presenter explains your software around your real screenshot, in 16:9 and 9:16Software text, screenshot, a voice
Tour from photos: homes, stays and venuesA narrated, captioned tour in 9:16 and 16:9Listing text, six photos, a voice
Local business promoA hosted promo of your place with your offer, captions and musicBusiness text, three photos
Finish a raw clipYour clip back with sound effects, music and captions, in 9:16 and 4:5One raw clip
  • The text boxes hold example copy about an invented product. Replace it with your brand facts, written out: the AI Writer steps cannot open a link.
  • A talking actor step takes its face from an earlier step and runs the two steps itself. Leave the actor empty, or the face is replaced.
  • The voice is matched to the face. Leave it empty and each new face gets a voice that fits it: a man's voice for a man, a woman's for a woman. Set a voice (a voc_ id) only to use the same one every run. If the face cannot be read, the run uses a default voice and still finishes.
  • A template with a voice-over step needs one voice. The voice-over cannot follow the face, so set the same voice on it and on each talking step.

Templates that need files

Five templates run from words alone: Creator to camera, Four openings, one ad, Two people: podcast, interview or skit, Long-form sales video and Sale ad in every size. A run cannot carry files, so for any other template, add them in the app first:

  1. Open the template in the app and drop the files into its upload cards.
  2. Click Open in Agent in the workflow's top bar, then Copy prompt. The prompt carries the saved workflow's id.
  3. Paste it to your agent. It runs that saved workflow, files included.

The first time an agent runs a template, it makes a copy in your workspace named <template name> (agent). Later runs reuse that copy, so files you add to it stay.

Run a workflow

List the templates, start a run, then read the run until it ends.

Tool calls
1. list_templates     no arguments
2. run_workflow       the arguments below
3. get_workflow_run   workflow_run_id, again until still_running is false
run_workflow arguments
{
  "template_key": "ugc_face_cam",
  "max_credits": 1500,
  "inputs": [
    {
      "node": "f_brand",
      "field": "text",
      "value": "Brand: Northpeak trail socks. ... Creator: a man in his thirties."
    },
    {
      "node": "f_video",
      "field": "voice",
      "value": "voc_0193c8f0a1b24e7f9d3c5a6b7e8f0099"
    }
  ]
}
get_workflow_run arguments
{
  "workflow_run_id": "wfr_..."
}

Send template_key or workflow_id, never both.

The start answers with a workflow_run_id (wfr_), not a finished ad. Names can change, so match templates on template_key.

Rules for every run

  • max_credits covers every step, not one. There is no estimate for a whole run.
  • Too low and the whole run is refused with max_credits_exceeded. limit.required_credits says what it needs, and nothing is charged.
  • Workspace limits apply: the per-generation limit to each step, the daily budgets to the whole run. See Spend limits.
  • One run per workflow at a time. A second start gets submission_in_flight (409).
  • Inputs set text, settings, actors and voices for this run only, never files or the number of takes. One bad input refuses the whole run.
  • Read the run about every 30 seconds until still_running is false. Files are in nodes[].generations[].outputs[]: Download links.
  • Agents cannot build, edit, stop or retry a workflow: do that in the app. On partial or failed, read each step's error, fix the input and start a new run.

Every field, status and error: Workflows API.

On this page