FlowModels

Generate videos: Playground, API, models and costs

Published on 13. November 2026

With FlowModels, you can generate short videos from a text description or from a start image. A video is created in the background: you start a job, we reserve the amount, and as soon as the video is ready, you can play or download it. You can do this in the Playground in the dashboard or directly through the API.

The models

Model Input Duration Resolution Sound
bytedance/seedance-2.5/text-to-video prompt 4 to 30 seconds 480p, 720p, 1080p yes
bytedance/seedance-2.5/image-to-video start image, prompt optional 4 to 30 seconds 480p, 720p, 1080p yes
minimax/h3/text-to-video prompt 5 to 15 seconds 2K no
minimax/h3/image-to-video start image and prompt 5 to 15 seconds 2K no
alibaba/wan-3.0/text-to-video prompt 2 to 30 seconds 480p, 720p, 1080p yes
alibaba/wan-3.0/image-to-video start image and prompt 2 to 30 seconds 480p, 720p, 1080p yes

All image-to-video models also accept an end image. Only use models that appear with the kind video in your response from GET /v1/models. There, video lists the allowed values per model and pricing.variants the prices per second. There's no automatic switch to another model.

In the Playground

Open the Playground in the dashboard and switch to the Videos tab. Choose a Model, describe the scene in the Prompt field and set Duration in seconds and Resolution. Depending on the model, Aspect ratio, Output format, Generate sound, Thinking mode, Add AI watermark or Seed (optional) are added. For image to video, you upload a Start image and, if you like, an End image (JPEG, PNG, WebP or GIF, up to 10 MB).

Before you start, Reservation shows the amount set aside for your selection and the price per second. Generate video starts the job.

The video Playground with the form, the reservation and the list of videos

Your videos lists all videos of your team, including the ones started through the API. While a video is running, the list refreshes on its own. You can play finished videos right there or download them, cancel a video that is still waiting and delete a finished one. The list doesn't show prompts because the whole team can see it. Billing runs through your team's FlowModels credit, so you don't need an API key for this. In Usage and Requests, your calls appear as "Playground · your name".

What the other Playground tabs can do and which limits apply there is explained in Try models in the Playground.

Via the API

Send the job to POST /v1/videos:

curl https://api.flowmodels.one/v1/videos \
  -H "Authorization: Bearer at_your_key" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: video-summer-campaign-01" \
  -d '{"model": "bytedance/seedance-2.5/text-to-video", "prompt": "A paper boat on a calm lake at sunrise", "duration": 5, "resolution": "720p"}'

The API responds right away with HTTP 202 and a job with an id and status queued. With the same Idempotency-Key, you can safely repeat the request after a network error, and no second video is created.

The most important fields:

  • model and prompt: The prompt is required for all text-to-video models and for MiniMax and Wan image to video, with up to 7,000 characters.
  • image_url or image_file_id: the start image as an https address, as a data: URL or as a file ID from POST /v1/files. For the end image, there are end_image_url and end_image_file_id.
  • duration, resolution, aspect_ratio: duration in seconds (default 5), resolution and aspect ratio within the range of the model. Seedance image to video follows the aspect ratio of the image.
  • generate_audio (Seedance and Wan), output_format with mp4 or mov (Seedance text to video only), enable_thinking and seed (Wan only), aigc_watermark (MiniMax only).

If the model doesn't know a field, the API rejects it with 422 and the field name in param before anything is reserved.

Status and retrieval

  1. Query the job with GET /v1/videos/{id}, at most every five seconds. Alternatively, a webhook reports the events video.completed, video.failed and video.cancelled, see Set up FlowModels webhooks.
  2. status is queued first, then in_progress, and finally completed, failed or cancelled.
  3. Once the video is ready, the response contains content_url, content_type and size_bytes. GET /v1/videos/{id}/content responds with a 302 to a download address that is valid for 15 minutes.

GET /v1/videos lists your team's jobs, and you can page and filter with limit, after and status.

With POST /v1/videos/{id}/cancel, you cancel a job as long as generation hasn't started yet, and the reservation is released. If it's already running, the API responds with 409 video_not_cancellable, and the video is completed and billed.

Retention

A finished video stays available for seven days after completion, and expires_at gives the exact time. After that, we delete the video and the job automatically. With DELETE /v1/videos/{id} or in the Playground, you delete a finished video earlier.

What a video costs

Videos are billed per second. The price depends on the model, the resolution, the sound and, for Wan, the thinking mode. You can see the current price under Models and in GET /v1/models under pricing.variants.

When the job starts, we fix the price and reserve the full requested length. The amount is only charged once the finished video is stored with us. Failed, rejected and cancelled jobs cost nothing. What was billed is shown in the finished job under usage.cost.

Limits

  • How many videos your team can run at the same time depends on your request limit tier: 3, 5, 10 or 20. See Request limit: how the tiers work.
  • Start and end images can be up to 10 MiB.
  • The generated videos are AI generated. Label them accordingly when you publish them.

If something goes wrong

  • 422 invalid_video_request: A field doesn't fit the model, and param names it.
  • 422 video_input_invalid: The image is missing, invalid, too large or doesn't belong to your team.
  • 429 insufficient_quota or 429 spend_limit_exceeded: Your credit or spending limit isn't enough for the reservation.
  • 429 video_concurrency_exceeded: Too many videos are already running. Wait for the time in Retry-After.
  • 503 video_model_unavailable, 503 video_pricing_unavailable or 503 video_provider_unavailable: Try again later, your credit wasn't charged.
  • 404 video_not_found: The ID is unknown or doesn't belong to your team.
  • 409 video_not_ready or 409 video_not_deletable: The current status doesn't allow the action yet.
  • Job failed: error.code gives the reason, for example video_generation_failed or video_content_rejected. In rare cases, such as video_submit_uncertain, the amount stays reserved until we have checked the case.

Was this article helpful?

Back to Models, limits & errors