Video

Video is a job. You get an id at once and poll for the result. The model is <provider>/<model>, for example xai/grok-imagine-video or openrouter/google/veo-3.1-fast.

Where Aile relays the download, the link in a completed job's status works without a key for an hour. Anyone holding it can download the video.

1curl https://api.aile.sh/v1/videos \
2-H "Authorization: Bearer $AILE_KEY" \
3-H "Content-Type: application/json" \
4-d '{ "model": "xai/grok-imagine-video", "prompt": "a lighthouse at dusk", "seconds": "6" }'
5
6curl https://api.aile.sh/v1/videos/$VIDEO_ID -H "Authorization: Bearer $AILE_KEY"

Billing

• You pay for the seconds you asked for, once the provider reports the video done. Lipsync and avatar models are the exception: you pay for the length produced (below). So is BytePlus Seedance: you pay the tokens it reports (below).
• xAI's Grok Imagine models and openrouter/heygen/heygen-video-1 are priced per resolution: you pay the per-second price of the resolution you send, or of 480p when you send none. A resolution the model does not list is a 400. Listings show the dearest resolution. HeyGen Video 1 renders 2K text-to-video at 16:9 or 9:16 only; any other aspect_ratio at 2K is a 400.
• A failed, empty or expired job is not charged. Jobs expire after 30 minutes by default.
• The hold stays on your balance while the job runs, and counts toward the key's spend limit (a 402 once open jobs hold the rest of it).
• A job id is readable only by keys of the account that created it.
• size ("1280x720") passes through where the model supports it.
• One input image is accepted as input_reference (JSON, or an uploaded file), image, or one first_frame in frame_images, as a data URL or an https URL. A text-to-video model refuses it with a 400.
• Other references (last_frame, reference_images, reference_audios, keyframes, video, previous_job_id, and input_references) are a 400 before anything is held, never silently dropped, except on the lipsync, avatar and Seedance models that take them.
• OpenRouter video models refuse a request for an audio track.
• Gemini's Veo (gemini/veo-3.1-generate-preview, veo-3.1-fast-generate-preview, veo-3.1-lite-generate-preview) is text-to-video only, 4, 6 or 8 seconds, with sound included in the price. Standard renders 720p or 1080p (1080p at 8 seconds only); Fast and Lite render 720p. The aspect ratio is 16:9 or 9:16. Anything else is a 400 before anything is held.
• When a sharer's provider account turns out to have no quota for a model, the request is a 502 saying you were not charged, and that sharer stops being offered that model; a retry reaches another sharer.
• A provider refusal you can fix (400, 404, 413, 422 …) passes through. One about the sharer's account is a 502 saying you were not charged. A 502 or 504 arrives as 503 with the real code in x-aile-status.
• A key's pins and the routing headers choose the sharer here too. x-aile-provider must match the model's prefix, and x-aile-max-price does not apply to a per-second price.
• Keyless x402 works here when the deployment enables it, except for lipsync, avatar and BytePlus Seedance models: you pay when the render starts, and a render that fails or expires is refunded to the wallet that paid. Poll the job by its id with no key.

Lipsync and avatar

Three sync.so models re-sync a video's lips to an audio track. openrouter/heygen/avatar-iv animates a photo to speak a script or an audio file, and lists at $0.05 per second. The length of the result comes from your media, so you name its length in seconds.

Lipsync is billed per frame produced. The prices above are at list.

1curl https://api.aile.sh/v1/videos \
2-H "Authorization: Bearer $AILE_KEY" \
3-H "Content-Type: application/json" \
4-d '{
5  "model": "sync/lipsync-2",
6  "seconds": 30,
7  "input_references": [
8    { "type": "video_url", "video_url": { "url": "https://example.com/clip.mp4" } },
9    { "type": "audio_url", "audio_url": { "url": "https://example.com/voice.wav" } }
10  ]
11}'

• seconds is required: a whole number from 1 to the model's longest (180 on avatar-iv). Name it at or above your media's length.
• On lipsync you pay for the frames in the finished video; if Aile cannot read the file, its length disagrees with the provider's, or it runs under 10 or over 240 frames a second, you pay the model's full per-second price for each second produced.
• sync-3 is slow: about 10 to 15 minutes for 30 seconds.
• A sharer on sync.so's free, Hobbyist or Hobby plan returns a watermarked video, and Aile cannot tell in advance.
• On avatar you pay for the length produced, rounded up to a second.
• Neither is ever more than seconds + 2% + 1 s (rounded up). A render longer than that is withheld and not charged. If no length can be found at all, you pay for the full seconds.
• At most three of these renders can be open per account at once, and repeats are rate limited: after three in an hour that were withheld or ran past the job's 30 minutes, the next create is a 429.
• video_url and audio_url must be https URLs the provider can fetch. Aile never fetches them. Send at most one of each kind.
• Lipsync takes a video and an audio track, and renders at the video's size, so size, resolution and aspect_ratio are a 400. sync_mode is bounce, loop, cut_off, silence or remap; cut_off when you send none. prompt is ignored on lipsync.
• Avatar takes one photo (an image_url reference, or input_reference) and either a script in prompt or an audio_url. voice_id picks the voice for a script.
• A 400 saying the output is longer than the sharer's plan renders means: send shorter media.
• These models need an API key. Keyless x402 answers 401.

BytePlus Seedance

ByteDance's Seedance models, served from a sharer's BytePlus ModelArk account.

• You pay the tokens BytePlus reports. The hold is the price above times seconds, priced at the widest frame for that resolution, so a 16:9 video costs a little less. Prices are at list.
• seconds defaults to 5. resolution defaults to 720p, or 1080p on the 1.0 models.
• Every model takes a first frame. The 2.5 and 2.0 models also take a last_frame, up to 30 (2.5) or 9 (2.0) reference images and up to 10 or 3 reference audio clips. 1.0 pro takes a last_frame only.
• Send references as input_references (image_url and audio_url items), reference_images or reference_audios. A frame and references cannot be mixed. A reference video is a 400.
• aspect_ratio is 16:9, 4:3, 1:1, 3:4, 9:16, 21:9 or adaptive.
• The 2.5 and 2.0 models add sound unless you send generate_audio: false, at the same price. The 1.0 models are silent.
• BytePlus refuses reference images that show real faces.
• These models need an API key. Keyless x402 answers 401.