Auto-Generate a Whole AI Campaign with One API

Short answer: platforms like WaveSpeed put 1,000+ image and video models — FLUX, Seedance, Kling, Veo, Z-Image Turbo and more — behind one unified REST API. You POST a prompt to a model's endpoint, get a task id back, poll a result URL until the status says completed, and download the output. Feed that loop a numbered shot list and a ~30-line script renders your AI model's entire campaign hands-free — no GPU, no installs, pay per image.

This is the third path we've covered: own GPU or rented GPU gives you control and privacy; the unified API gives you zero setup and every model on one bill. Here's the exact mechanics.

When the API path wins

• Zero setup. No drivers, no 20 GB model downloads, no VRAM math. Your first image is one HTTP request away.
• Every model, one bill. Open-weight models your laptop can't hold and hosted video models (Seedance, Kling, Veo) live behind the same auth header.
• Honest per-image pricing. As a reference: Z-Image Turbo starts around $0.005 per image at base settings (~200 images per dollar); FLUX.2 Dev around $0.025. The exact price for your parameters is shown on each model page before you submit — outputs, sizes and options change the cost.
• The trade-offs: you pay per run, you need to be online, and the provider's content policies apply — this is a hosted service, not your private machine. For brand-catalog fashion work that's usually fine; know the rules of what you're rendering (see our swimwear & lingerie guide).

The API in sixty seconds

Three verified moves (from the official docs):

1. Submit: POST https://api.wavespeed.ai/api/v3/{provider}/{model} — for example wavespeed-ai/z-image/turbo — with headers Authorization: Bearer <API key> and Content-Type: application/json, body containing your prompt and options. The response returns a task id.
2. Poll: GET /api/v3/predictions/{id}/result — status goes created → processing → completed (or failed with an error field). Poll every ~2 s for images, ~5 s for video, backing off for long jobs; webhooks exist if you'd rather be notified.
3. Download: when status is completed, the file URLs are in the outputs array. Save them with the shot's number so the campaign stays organized.

The whole loop, in pseudocode

jobs = load("shot_list.json")   # numbered prompts, identity locked
for job in jobs:
    r = POST(f"https://api.wavespeed.ai/api/v3/{MODEL}",
             headers={"Authorization": f"Bearer {KEY}"},
             json={"prompt": job.prompt, **job.options})
    task_id = r.data.id
    while True:
        res = GET(f".../api/v3/predictions/{task_id}/result")
        if res.data.status == "completed": break
        if res.data.status == "failed": log(res.data.error); break
        sleep(2)   # ~2 s images, ~5 s video
    save(res.data.outputs[0], f"shot_{job.number}.png")

That's the entire automation. Run it, make coffee, come back to a folder of finished shots.

Let the AI drive it

You don't even have to write that script yourself. Paste your shot list and identity brief into ChatGPT or Claude and ask it to produce the runner for your chosen model — then to review the outputs against the identity and flag any shot that drifted. If you keep your model in a project folder, the assistant already has the identity and the campaign on hand; the API key is the only new ingredient. Never paste the key into anything you don't control.

Choosing the model

• Photoreal stills on a budget: Z-Image Turbo — excellent skin realism at the cheapest tier.
• Maximum image control: the FLUX family — pricier per image, strong prompt adherence.
• Video: Seedance, Kling and Veo endpoints take the same submit-and-poll loop; prices are per model and per duration, so read the model page first.
• Rule of thumb: iterate on one shot until the recipe is right, then batch the rest. A 21-shot campaign on Z-Image Turbo at base settings costs on the order of a dime — mistakes are cheap, but they're cheaper still when you dial the prompt in first.

Keep the identity locked

An API doesn't change the consistency rules: the same identity tokens in every prompt, the same reference where the model supports it. Batch generation actually rewards discipline — one clean identity block, copied into 21 prompts by a script, drifts less than 21 hand-typed variations ever would. The groundwork is in keeping an AI character consistent.

Where the shot list comes from

Any numbered prompt list works. The GoldenPrompts AI Model Studio exports exactly this shape: a jobs file with your model's identity baked into every campaign prompt, plus a master brief that tells ChatGPT or Claude how to run the batch and validate every output. Click the campaign together, download the pack, point the loop at it.

FAQ

Want the shot list without the typing? The AI Model Studio builds your model's identity, campaign prompts and content pack in a few clicks. Free to start: 24 hours of everything, no card.