AI Video for Ads via API: Text-to-Video & Image-to-Video Clips

A performance marketing team testing ten hooks across three aspect ratios needs thirty clips, not one hero video — a budget and timeline no production shoot fits. Aigenway puts Kling, Veo and Seedance behind the same provider/model/method contract, so generating ad variants from a text brief or an existing product still is a script you run, not a shoot you book.

What an ad pipeline needs from a video API

Ad clips have constraints a general demo reel doesn't: a fixed duration to match a placement, a specific aspect_ratio per platform (9:16 for Reels and TikTok, 1:1 for feed, 16:9 for YouTube pre-roll), and a turnaround fast enough to iterate on a hook before the campaign launches. Every model on the gateway takes the same duration and aspect_ratio params regardless of provider, so the constraint lives in your request, not in provider-specific glue code. Confirm the live model codes and per-second pricing on your account before you build against one:

curl "https://api.aigenway.com/v1/methods?provider=kling" \
  -H "Authorization: Bearer sk_live_..."

Don't hardcode a version string from this post — providers rev their lineup on their own schedule, and /v1/methods is the ground truth for what your account can actually call.

Animate a product still into a vertical ad

Most ad creative starts with an asset you already have: a product photo, a packshot, a brand still. image_to_video takes that image and a motion prompt and returns a short clip framed for the placement you need — no reshoot, no stock footage:

curl -X POST https://api.aigenway.com/v1/generations \
  -H "Authorization: Bearer sk_live_..." \
  -H "Idempotency-Key: $(uuidgen)" \
  -H "Content-Type: application/json" \
  -d '{
    "provider": "kling",
    "model": "kling-v2-master",
    "method": "image_to_video",
    "params": {
      "prompt": "slow push-in on the product, soft studio light shifting, subtle steam rising, cinematic ad look",
      "input_image": "https://cdn.example.com/products/bottle-still.jpg",
      "duration": 5,
      "aspect_ratio": "9:16"
    }
  }'

The response is a task, not a finished file — video render is never instant:

{
  "task_id": "cmomx7b3c000scr04h2q9d81f",
  "status": "pending",
  "provider": "kling",
  "model": "kling-v2-master",
  "method": "image_to_video",
  "created_at": "2026-09-29T09:12:03.000Z"
}

Generate a concept spot from just a brief

When there's no still to start from — a new product, an abstract concept, a scene nobody has shot — text_to_video builds the clip directly from a prompt. Same endpoint, same task contract, just a different method and no input_image:

curl -X POST https://api.aigenway.com/v1/generations \
  -H "Authorization: Bearer sk_live_..." \
  -H "Idempotency-Key: $(uuidgen)" \
  -H "Content-Type: application/json" \
  -d '{
    "provider": "seedance",
    "model": "seedance-1-pro",
    "method": "text_to_video",
    "params": {
      "prompt": "a runner laces up at dawn on an empty city street, energetic, handheld camera feel, product logo visible on the shoe",
      "duration": 5,
      "aspect_ratio": "16:9"
    }
  }'

Seedance's lower cost-per-second makes it a reasonable default for early-concept passes where you're testing whether an idea works at all before committing to a higher-fidelity render on Kling or Veo — switching between them later is a one-line change to provider/model, not a rewrite.

Batch hook and aspect-ratio variants for A/B testing

Ad testing lives on variants: the same core idea across several hooks, several first frames, several crop ratios. Loop the same params shape over the axis you're testing and give each submit its own Idempotency-Key so a retry on a flaky connection can't double-charge a variant:

for ratio in "9:16" "1:1" "16:9"; do
  curl -X POST https://api.aigenway.com/v1/generations \
    -H "Authorization: Bearer sk_live_..." \
    -H "Idempotency-Key: $(uuidgen)" \
    -H "Content-Type: application/json" \
    -d "{
      \"provider\": \"kling\",
      \"model\": \"kling-v2-master\",
      \"method\": \"image_to_video\",
      \"params\": {
        \"prompt\": \"slow push-in on the product, soft studio light shifting, cinematic ad look\",
        \"input_image\": \"https://cdn.example.com/products/bottle-still.jpg\",
        \"duration\": 5,
        \"aspect_ratio\": \"${ratio}\"
      }
    }"
done

Swap the loop variable for a hook line embedded in the prompt instead of the aspect ratio, or nest both, and the same pattern produces a full test matrix from one product still — the identical batching approach used for product photo variants applies directly to ad creative.

Collect finished clips by webhook, not by polling

A real test matrix is dozens of renders running concurrently — polling each one is wasted requests and added latency. Pass a callback_url on every submit and let the gateway push the finished clip to your endpoint the moment it's ready:

curl -X POST https://api.aigenway.com/v1/generations \
  -H "Authorization: Bearer sk_live_..." \
  -H "Idempotency-Key: $(uuidgen)" \
  -H "Content-Type: application/json" \
  -d '{
    "provider": "veo",
    "model": "veo-3",
    "method": "text_to_video",
    "params": {
      "prompt": "product hero shot, bright kitchen counter, morning light through a window",
      "duration": 5,
      "aspect_ratio": "16:9"
    },
    "callback_url": "https://your-app.example.com/webhooks/aigenway"
  }'

Each delivery carries an X-Aggregator-Task-Id you map back to the variant that requested it, and an X-Aggregator-Signature HMAC to verify before you pull the clip into your ad account. The webhook payload shape is identical no matter which of the three models rendered it — see the webhooks docs for the full contract and the signature check.

Why one gateway beats separate Kling, Veo and Seedance accounts

• One contract across three models. text_to_video and image_to_video take the same params shape whether the provider is Kling, Veo or Seedance — testing all three on one brief is a config change, not three integrations.
• Match cost to the stage. Run cheap, fast Seedance passes for early concept testing, then re-render the winning hook on Kling or Veo for the version that actually ships.
• Prepaid USD billing, one balance. No separate credit pack per provider to forecast against a campaign budget — check GET /v1/balance before queuing a large test matrix.
• Idempotency and webhooks built in. Retry-safe submits and signed callbacks come with the task contract, not as something you bolt on per vendor.

Shoot or generate the still once, loop the variants you actually need to test, and let the webhook — not a polling loop — tell you when the matrix is ready to review.