apimart-zoe API

apimart-zoe is a multimodal video generation model from APIMart, called with the model ID apimart-zoe. The apimart-zoe API supports text to video, first/last-frame video, reference to video with images, videos and audio, editing of an existing video, optional audio generated together with the video, and a two-step 480p draft to 1080p final workflow. Clips run 4–30 seconds at 480p, 720p, 1080p, 2k or 4k.

Send POST /v1/videos/generations with model apimart-zoe and your prompt, using Authorization: Bearer <API key>. The response returns a task_id. Generation usually takes 1–3 minutes: poll GET /v1/tasks/{task_id} every 5–10 seconds until data.status is completed, then read the video URL from data.result.videos[0].url[0]. The example on this page has cURL and Python code, and the full apimart-zoe API reference is in the APIMart docs.

model (apimart-zoe) is required. prompt is required for text to video, optional when you only send media, and can be up to 204,800 characters. Optional parameters are resolution (480p, 720p, 1080p, 2k or 4k), aspect_ratio or size (21:9, 16:9, 4:3, 1:1, 3:4, 9:16 or adaptive), duration or seconds (4–30 or -1), audio or generate_audio, seed, image_with_roles, image_urls (aliases image, images, image_url), first_frame_image, last_frame_image, video_urls, audio_urls, draft, draft_task_id, omni_reference_task_type (auto, reference or edit), return_last_frame, output_format or response_format (mp4 or mov) and enhance_prompt, plus the APIMart field nsfw_check. negative_prompt, camera_fixed (including camerafixed), extend_from_task_id, mask_url, video_list, element_list, multi_shot and tools are not supported: if the field appears in the request at all, even empty, false or null, it returns 400. n can only be omitted or set to 1. The parameter table on this page lists defaults and limits.

Set duration (or seconds) to a whole number from 4 to 30; the default is 5, and -1 lets the model decide. resolution is 480p, 720p (default), 1080p, 2k or 4k and is case-insensitive. aspect_ratio, also accepted as size, is 21:9, 16:9, 4:3, 1:1, 3:4, 9:16 or adaptive (default), where the model or your media sets the frame. For a video edit, leave out duration and aspect_ratio (or send -1 and adaptive) and the clip follows the source video's length and frame. Drafts are always 480p and finals 1080p.

Yes. Send image_with_roles with first_frame and last_frame, or use first_frame_image and last_frame_image. Each frame takes one image, and a last frame needs a first frame. A first/last-frame request cannot also include reference images, videos or audio; such a request returns 400 and is not charged. Use the prompt to describe how the shot moves from one frame to the other.

Up to 30 reference images, 10 reference videos and 10 reference audio clips, 50 files in total; reference audio must come with a reference image or video. Images can be jpeg, png, webp, bmp, tiff, gif, heic or heif, 300–6000 pixels per side, aspect ratio 0.4–2.5 and under 30 MB. Videos are mp4 or mov (H.264/H.265, with an AAC or MP3 audio track), 2–15 seconds each (1.8–15.2 s is accepted) and up to 200 MB, 300–6000 pixels in width and height with width × height between 409,600 and 8,295,044 (about 640×640 to 3840×2160), aspect ratio 0.4–2.5 and 24–60 fps after rounding (23.976 and 59.94 are accepted); the source video of a video edit can be 4–30 seconds (at least 3.8 s and under 30.2 s), and without a task type, videos longer than 15 seconds need edit. Audio is wav or mp3, 2–15 seconds (1.8–15.2 s) and up to 15 MB each, 30 seconds in total (up to 30.2 s). Reference video length, dimensions, aspect ratio and frame rate, and reference audio length, are checked at submission: a file that does not fit returns 400 invalid_request_error and is not charged. Image dimensions, file size and encoding are checked during generation: a file that does not fit fails the task and the charge is refunded automatically, so check your files before calling. All media must be public http(s) URLs or asset:// IDs; data: URIs are not accepted, so upload files to a public address first. On submission APIMart registers each file automatically, and the same file reused within 7 days is not registered again. When you only send media, prompt can be left out.

Submit a request with draft: true to get a low-cost 480p preview; drafts can only be 480p and cannot be combined with a video edit. When the draft succeeds, submit a second request with model and draft_task_id set to the draft's task_id. The final is rendered at 1080p and reuses the draft's prompt, media, duration, aspect ratio, seed and audio settings. Apart from model, draft_task_id, resolution and common fields such as nsfw_check or webhook, no field may be sent: prompt, media, duration, aspect_ratio, seed, audio, omni_reference_task_type, return_last_frame, output_format or enhance_prompt returns 400, even when empty. Create the final within 24 hours of the draft: a draft_task_id that does not exist, is not a draft, is unfinished or is older than 24 hours returns 400. The draft and the final are each billed as a generation; the final is billed at 1080p for the draft's duration, or for the draft's actual length if the draft used -1.

Submission errors return an error object with message, type, param and code, and are not charged. 400 invalid_request_error covers invalid values, media counts and combinations (such as first/last frames with reference media, or image_urls with image_with_roles), reference video length, dimensions, aspect ratio or frame rate and reference audio length out of range, task type and video edit rules, output_format values, draft and final rules, and unsupported parameters such as negative_prompt or camera_fixed. 400 input_video_probe_failed means a reference video cannot be reached or its length cannot be read. 402 payment_required means the balance is too low; with duration -1 or a video edit, 30 seconds are reserved. 400 nsfw_content_detected means nsfw_check flagged the prompt or images; message lists the matched categories. 503 draft_channel_unavailable or draft_task_unavailable means drafts are temporarily unavailable; retry later or create a new draft.

A clip usually takes 1–3 minutes, so poll every 5–10 seconds. data.status is pending, processing, completed or failed, and data.progress runs from 0 to 100. The video URL in data.result.videos[0].url is stored on APIMart and expires 72 hours after completion, so download it in time. With return_last_frame: true, data.result.last_frame_url also returns the last frame as a png image. If a task fails, including when media registration does not pass or the image dimension, file size or encoding check fails during generation, data.error.message gives the reason, data.error.code is task_failed, and data.cost is 0 because the reserved amount is refunded automatically.

The apimart-zoe API is priced per output second, with rates for 480p, 720p, 1080p, 2k and 4k and a separate rate for requests with a reference video. If you set duration (4–30) or leave it at the default 5, you are billed for the requested seconds, without adjustment to the finished clip. With duration -1 or a video edit (omni_reference_task_type: edit), 30 seconds are precharged and, after completion, settled on the clip's actual length rounded to the nearest second; a video edit uses the rate for requests with a reference video. With a reference video, billed seconds = video duration + max(total reference video length, ⌈video duration × 2/3⌉); the reference length is measured per video at submission, summed, rounded down to whole seconds and not adjusted later. A final made from a draft is billed at 1080p for the draft's duration (the draft's actual length if the draft used -1) and, if the draft used a reference video, also counts the reference video length. Your charge reflects your membership or account group discount; see the live pricing table on this page.

On APIMart both models use POST /v1/videos/generations and GET /v1/tasks/{task_id}, create 4–30 second clips in aspect ratios from 21:9 to 9:16, accept up to 30 reference images, 10 videos and 10 audio clips per request, support seed, audio generation, return_last_frame, output_format and video edit (omni_reference_task_type: edit), and offer a 480p draft to 1080p final workflow. The differences: apimart-zoe (model ID apimart-zoe) also outputs 2k and 4k, does not allow first/last frames together with reference media, needs a reference image or video whenever you send reference audio, and lets the model decide on audio when audio is omitted, while Seedance 2.5 (seedance-2.5) turns generate_audio on by default and also supports extend, which apimart-zoe has not opened yet. With a reference video, the two models also count billed seconds differently. The comparison tables on this page list each difference and the current per-second prices.

It depends on the resolution and on whether you send a reference video. The price comparison on this page shows both models' current per-second rates with and without a reference video, the difference in percent, and cost examples for common clip lengths. With a reference video, apimart-zoe adds max(total reference video seconds rounded down, ⌈output seconds × 2/3⌉) to the output seconds, while Seedance 2.5 adds the reference video seconds, so compare the example that matches your clip. apimart-zoe also has 2k and 4k tiers that Seedance 2.5 does not offer. Rates are read from the APIMart pricing API and depend on your membership tier.

No. It is pay-as-you-go and billed per output second, and a draft and its final are each charged as a generation. Submission errors are not charged, a request returns 402 if your balance is too low, and failed tasks are refunded automatically, with data.cost showing 0. Your price depends on your membership tier or account group; the live pricing table on this page shows the current rate for each resolution.

Sign up for APIMart, open the API Keys page in your dashboard and create a key, then send it as Authorization: Bearer <key>. The same key calls the apimart-zoe API and the other models on APIMart, with pay-as-you-go billing and no subscription.