Seedance 2.0@ext Short-Form Video: A Practical API Workflow for Social Clips

Short-form video is where most social discovery happens now, and a clean vertical clip usually means a shoot, an editor, or both. seedance-2.0@ext on hiapi skips that: write one detailed prompt, call an API endpoint, and get back a finished 9:16 video — audio included — a couple of minutes later.

This walkthrough covers the workflow end to end using a real generated clip: how the prompt was structured, how the async task API call actually works, what it cost, and where the pricing stops behaving the way you'd expect.

Why seedance-2.0@ext for Short-Form Clips

seedance-2.0@ext is the ext route of the seedance-2.0 model family on hiapi — a separate route with its own pricing and its own parameter set, distinct from the default seedance-2.0 route. It's built specifically around vertical, social-native output: native 9:16 framing, camera moves that read well on a phone screen, and audio generated in the same pass as the video instead of bolted on afterward.

That combination — vertical framing plus native audio — is what makes it a fit for TikTok- and Reels-style clips specifically, versus a general-purpose video model where you'd crop and score the audio yourself after the fact.

The Prompt: What Actually Went Into This Clip

We generated one demo clip end to end for this piece: a vertical product shot of cold brew coffee pouring over ice in a glass tumbler, styled for a social feed. Here's the exact output, followed by the exact prompt that produced it.

Vertical 9:16 close-up product showcase for a social feed: a frosted glass tumbler sits centered on a sunlit marble countertop. Camera starts with a slow overhead top-down push-in on the empty tumbler filled with clear ice cubes catching golden hour light. At two seconds, a stream of cold brew coffee pours in from off-frame top, swirling and darkening the ice, cream ribboning in gentle spirals just after. Condensation beads run down the glass as cool vapor drifts upward in backlight. Camera performs a slow arc from top-down to a low three-quarter hero angle, ending on a crisp close-up with shallow depth of field and warm bokeh from a window behind. Bright natural commercial lighting, glossy reflections, shallow depth of field, ultra clean product-video aesthetic, smooth handheld micro-motion, premium commercial finish. Native audio: soft pour trickle, ice clinking against glass, a light ambient cafe hush, no music. No readable text, no logos, no people, no hands, no cuts.

The result: 720×1280 (9:16), 24fps, roughly 8 seconds, with native AAC stereo audio baked into the file — no separate sound-design pass.

Why This Prompt Works

A few things this prompt does that a shorter, vaguer one wouldn't:

• It gives the camera a path, not a single shot. "Top-down push-in" → "slow arc to a low three-quarter hero angle" → "crisp close-up" describes movement across the whole clip, not one static composition.
• It times the action. "At two seconds, a stream of cold brew coffee pours in" anchors the pour to a specific moment instead of leaving it to guess when the "interesting part" should happen.
• It stacks material and lighting cues. Condensation beads, cool vapor, warm bokeh, glossy reflections — these are the details that separate a generic product shot from one that reads as premium.
• Audio is written directly into the prompt, not left as an afterthought: pour trickle, ice clinking, ambient hush, explicitly no music. The model syncs sound to the visual action described above it.
• It closes with a negative list. "No readable text, no logos, no people, no hands, no cuts" heads off the most common failure modes for product-style generations before they happen.

Treat the prompt as a short creative brief, not a caption — that's the actual difference in output quality.

Reproducing This With the /v1/tasks API

seedance-2.0@ext runs through hiapi's async task endpoint. You submit a job, poll it until it finishes, then pull the video from the response.

Submit:

curl -X POST https://api.hiapi.ai/v1/tasks \
  -H "Authorization: Bearer $HIAPI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "seedance-2.0@ext",
    "input": {
      "prompt": "<your prompt>",
      "aspect_ratio": "9:16",
      "resolution": "720p",
      "duration": 8,
      "generate_audio": true
    }
  }'

prompt and aspect_ratio are required — omit aspect_ratio and the call fails validation before it ever reaches generation. resolution, duration (4–15 seconds), and generate_audio are optional; the values above are close to what produced the clip in this article.

One thing worth flagging if you're coming from an image-to-video mindset: the pricing page lists an image-to-video rate for this model, but the live input schema only accepts a text prompt — image or reference-frame fields are rejected. This is a text-to-video model in practice, whatever the pricing tags suggest.

Poll for the result:

curl https://api.hiapi.ai/v1/tasks/<task_id> \
  -H "Authorization: Bearer $HIAPI_API_KEY"

Once status reports success, output[0].url points to the finished .mp4. That link is temporary — roughly a week based on the expiry window on the task response — so download the bytes and store them yourself immediately rather than linking to it directly.

Pricing: The 4K Quirk

seedance-2.0@ext prices per second of output, and the rate changes by resolution tier:

Resolution	Price per second
720p	$0.149
1080p	$0.373
2K	$0.662
4K	$0.467

Notice that 4K is cheaper than 2K on this model — $0.467/s versus $0.662/s. That's not a typo, and it's not the usual pattern; on most models here, price climbs monotonically with resolution. Don't assume it holds for other models, and re-check /api/pricing before you build resolution-based cost logic into a pipeline, since tiers like this do get adjusted.

For reference, the 8-second, 720p demo clip above cost $1.19 to generate — straightforward duration × per-second rate math at the 720p tier.

Choosing a Resolution for Where the Clip Actually Runs

Given the 720p → 1080p jump alone is roughly a 2.5x cost increase, resolution is the lever worth thinking about before you batch-generate anything:

• Feed and Stories placements (TikTok, Reels, Shorts) are almost always viewed on a phone at a fraction of 1080p's actual pixel density — 720p is generally indistinguishable there and is the cheapest tier.
• 1080p is worth it when the clip might get re-purposed for a landing page hero or a paid placement that could render larger than a phone screen.
• 2K and 4K make sense for source footage you plan to crop, zoom, or stabilize in post — extra resolution gives you room to reframe without visible quality loss. Given the pricing table above, 4K is the more efficient choice of the two on this specific model.

FAQ

Does seedance-2.0@ext support image-to-video, not just text-to-video? No — despite the pricing page carrying an image-to-video rate for this model, the live /v1/tasks schema for the ext route only accepts a text prompt; image URLs, reference frames, and reference-video fields are all rejected. Treat it as text-to-video only until the platform's model list says otherwise.

Can I control how long the generated clip is? Yes, via the optional duration field, an integer from 4 to 15 seconds. Leave it out and the model applies a default; setting it explicitly is safer if a downstream template expects a specific length.

Does the video come with audio automatically? Native audio generation is available through the optional generate_audio field. Writing the audio you want directly into the prompt — as with "soft pour trickle, ice clinking against glass, a light ambient cafe hush" above — gets it synced to the on-screen action without a separate sound pass.

Is 4K really cheaper than 2K, or is that a display bug? It's the current live pricing as of this writing (August 2026), not a bug — but resolution pricing can change, so confirm against /api/pricing before relying on it in a cost model.

How long is the returned video URL valid? The output[0].url from a completed task is a temporary link, valid for roughly a week based on the task's expiry metadata. Download and store the file yourself rather than linking to it long-term.

Where to Go From Here

Full parameter details and current pricing for this model live on the seedance-2.0@ext model page and the hiapi pricing page. For more prompt patterns built around this same model, Seedance 2.0@ext Prompt Recipes has additional worked examples, and a short-video product showcase built on Seedance 2.0 walks through the same async task workflow applied to a different creative brief.