# apicove.ai > apicove.ai is an OpenAI-compatible API gateway for generative media: image models, video models, Midjourney actions and audio. One bearer token, USD pay-as-you-go, no subscription tiers. Generated artifacts are served from our own edge and kept for 30 days instead of 24 hours. Base URL: `https://api.apicove.ai/v1` Authentication: `Authorization: Bearer $APICOVE_KEY` Docs: https://apicove.ai/docs ## What it does - Image generation: Gemini (Nano Banana), GPT-Image, FLUX 2, Qwen-Image, Seedream, MiniMax H3 image - Video generation: Seedance 2.x, Wan 3.0, SD2.x, MiniMax H3, Muse Video - Midjourney: imagine, blend, describe, upscale, variation, zoom, pan, inpaint, remix - Audio: Suno, FlowMusic ## Pricing All prices are per unit, in USD, and are the amount actually charged (no separate discount step). - Image models: priced per generated image - Video models: priced per second of generated output, so a 5-second clip costs 5x the per-second price - Midjourney: priced per action - Audio: priced per call Full table: https://apicove.ai/#pricing Machine-readable: https://apicove.ai/pricing.json ## Endpoints - `POST /v1/images/generations` — image generation, OpenAI-compatible request body. Usually returns `data[0].url`; some upstreams return `data[0].task_id` instead, so check for `task_id` first. - `POST /v1/videos/generations` — video generation, always asynchronous; returns `data[0].task_id` - `GET /v1/tasks/{task_id}` — task status at `data.status`, failure reason at `data.error.message` - `GET /v1/tasks/{task_id}/artifacts/{image|video}/content` — download the produced file - `GET /v1/models` — models enabled for the calling key - `GET /api/pricing` — all models with unit prices - `GET /api/status` — gateway status and configuration ## Artifacts Generation outputs are mirrored to our own object storage and exposed on `https://dl.apicove.ai/...`. Files are retained for 30 days, after which a lifecycle rule deletes them. Copy anything you need to keep before then. ## Notes for agents - Poll `/v1/tasks/{task_id}` every 3–5 seconds; polling faster returns nothing new. - Video endpoints take `duration` in seconds; unsupported values return HTTP 400 with a reason. - If a request returns 402, the balance is exhausted — there is no retry that fixes it. - Errors use one envelope: `{"error":{"code":"...","message":"...","type":"..."}}`. Branch on `code`, not on the message text. ## Terms https://apicove.ai/terms — prepaid balance, refunds within 14 days on unused balance, 30-day artifact retention. ## Contact hello@apicove.ai