SpicyAPI

Yes, at the platform level: we add no content classifier, no per-request declaration and no approval queue. Output still depends on the model you pick — Spicy editions are built for unrestricted generation, while mainstream models keep their own rules. Each model page tells you which is which.

Start with the Spicy editions — the model name ends in “Spicy” — and check the catalogue for the tasks each one covers. Several mainstream image and video models also follow uncensored prompts without adding a filter of their own; every model page states what it accepts. For image-to-video, begin from a still you like: the first frame steers the character and composition.

Prices are in US dollars, in each model's own unit — per image, per second of video or per million tokens. The Playground shows the estimate for your exact settings before you submit, and any active promotion shows the original and discounted price side by side.

Failed or timed-out tasks are usually not charged. Some models charge for content refusals as stated on their model page; check those billing terms before submitting. We reserve the estimated cost when a task is created, settle on actual usage when it finishes and release the difference. If a network error makes you resend, reuse the same Idempotency-Key within 24 hours so you never get a duplicate task or charge; use a new key for a new generation.

For chat, yes: point any OpenAI client at our base URL, swap in your key and a model name, and streaming works as-is. Video also offers the /v1/videos endpoint in the OpenAI Videos format. Image and video tasks use createTask, then recordInfo polling or a callBackUrl.

Yes. Pair an OpenAI-compatible LLM for in-character dialogue with image and video models for portraits and short clips, all on one key and one balance. To keep a character consistent, reuse the same reference image as the first frame or reference input for every shot.

Generated media is served from our own CDN, uploaded references are stored separately, and prompt text runs on a clock of its own. Our published maximums are 14 days, 1 day and 30 days — and each of those three is yours to wind shorter, for the whole account or for a single request. Copy anything you want to keep to your own storage before its clock runs out.

Yes. One call destroys a finished task's outputs, its stored result and its prompt text. It is scheduled the moment you ask, needs nobody's approval, and calling it twice is not an error. The billing record stays — the amount, the date, the model and the request ID — so a charge you might one day query is still there to look at. The task is then marked as destroyed by you, never mixed up with one that simply expired.