A free text to video API usually means trial credits, a watermarked tier, open weights on your own GPU, or a launch promotion. See real per second rates, the cost of a five second clip, the limits that bite, and a working PicassoIA Video API request.
Search for "text to video API free" and two kinds of results come back: listicles that promise unlimited clips at no cost, and pricing pages with per second rates that sting. The honest answer sits between them. A genuinely free text to video API exists in only a handful of forms, and each one has a catch worth knowing before you write a single line of integration code. This article separates the real options from the marketing, shows what a five second clip costs on the main providers, and walks through a working request on PicassoIA Video, whose API predictions are free right now for Infinite plan accounts.
Every "free" video API is a free sample, not a free meal. The vendor wants you to taste the output, hit a limit, and then pick a paid plan. That is not a scam, it is how the market works, but the sample comes in four distinct shapes and each one suits a different job.
Most hosted providers hand new accounts a small balance. It buys a handful of test clips, then the meter starts running. Treat those credits as a way to compare output quality across models, not as a production budget. A few 720p generations from a premium model can drain a starter balance in one afternoon, so decide which prompts you want to test before you spend anything. A simple plan works well: pick three prompts, run each one on every candidate model, and record the cost and the number of clips worth keeping in a spreadsheet.
Consumer video tools often give a daily allowance of free renders. The catch is a watermark and a resolution ceiling, and the allowance changes by account and region. PixVerse V6 is a typical example: according to one provider write up, free renders carry a watermark and top out around 540p. That is fine for checking whether a prompt works. It is useless for a client deliverable.
Some labs release the model files, so the software itself costs nothing. LTX Video, Hunyuan Video and Mochi 1 all ship open weights. What you pay for is the machine: buying a GPU with plenty of memory, or renting one by the hour. For a developer who already owns the hardware, this is the closest thing to a truly free text to video API. For everyone else it is a different bill, not a smaller one. Expect to spend an evening on drivers, memory settings and job queueing before the first clip renders. That setup time is the hidden price.
Providers sometimes switch billing off on a new endpoint to attract developers. PicassoIA does this today: its API page states that predictions are currently free and use no credits, while creating them requires an Infinite plan. Promotions end, so keep your integration behind a thin wrapper that lets you swap providers without rewriting your app.
💡 Rule of thumb: if a provider promises unlimited video with no card, no watermark and no rate limit, at least one of those three claims is false.
Paid text to video APIs bill by the second of output. The rate depends on the model, the resolution, and sometimes on whether the clip includes audio. The table below uses public rates reported by pricing trackers and provider listings at the time of writing. They move often, so confirm on the provider's own pricing page before you budget.
The formula is simple: seconds × rate × attempts. The last term is the one people forget. Video models are not predictable, so a prompt that lands on the first try is the exception. The number that matters is the cost per usable clip, not the cost per generated clip. If one generation in three is good enough to ship, multiply the sticker price by three.
Resolution moves the rate too. Many providers charge noticeably more at 1080p than at 720p, and a few add a surcharge when the request includes an input video. According to pricing trackers, Veo is not part of the Gemini API free tier, so every generation is billed.
Audio adds a wrinkle. Some models include the audio track in the per second rate, while others price it separately. Check which applies before you compare two rate cards, or the cheaper one may turn out to be the pricier one.
| Model and provider | Setting | Approx. rate per second | Five second clip |
|---|---|---|---|
| Veo 3.1 Lite via Gemini API | 720p | $0.05 | about $0.25 |
| Veo 3.1 Fast via Gemini API | 720p | $0.10 | about $0.50 |
| Veo 3.1 via Gemini API | 720p | $0.40 | about $2.00 |
| Wan 2.7 via fal | 720p | $0.10 | about $0.50 |
| PixVerse V6 via a third party host | 720p with audio | listed per clip | about $0.30 |
| PicassoIA Video via PicassoIA API | 480p to 1080p | $0 while the promotion lasts | $0 |
Here is what that means in practice. A twenty clip storyboard at $0.50 per clip, with three attempts for every clip you keep, comes to $30. The same storyboard at a $0.05 rate comes to $3. Cheaper models are not always worse, so run your own prompts through two or three of them before you commit to one.
💡 Tip: rates change when a newer model ships. Re-check the numbers at the start of every project instead of copying last quarter's budget.
No single option wins on every axis. These are the ones worth a test run, split by how much control you want over the hardware.
If you would rather not own a GPU, the same models sit in the PicassoIA text to video collection, where you can test a prompt in the browser before you commit to hosting anything.
For API work, the lowest hosted rates usually come from lighter or older models. These are the ones to benchmark first:
My ranking, for what it is worth: the PicassoIA API wins on price today, open weights win on control, and Veo 3.1 Lite wins on a predictable rate card from a large vendor.
The PicassoIA API allows up to 5 predictions queued or running at once, shared across your API credentials and MCP connections. That is a sensible ceiling for a prototype and a hard one for a batch pipeline. Other providers set their own caps, often lower on trial accounts. Build your job runner around a queue from day one: submit up to the cap, poll, and refill as slots free up.
Limits trade off against each other. On PicassoIA Video, 480p runs up to 20 seconds, 720p up to 10 seconds and 1080p up to 5 seconds. Higher resolution buys detail and costs length. Seedance 2.5 Lite, which adds synchronized audio, offers 5, 10 or 15 seconds at 480p or 720p.
Short clips at high resolution mean stitching in an editor. Plan for that in your timeline, because a ten second scene at 1080p becomes two generations, and that doubles the attempts term in your cost formula.
Free output is not always free to publish. Check three things before a clip goes into a client's ad:
Write the answers into your project notes. They decide whether a cheap clip can ship.
PicassoIA Video is the quickest route to a text to video API call with no per second charge today. The same API also serves Seedance 2.5 Lite for clips with audio. The flow is asynchronous: you create a prediction, poll it, and read the output once the status says succeeded.
pia_sk_, and each account can hold two. Store it in an environment variable, never in your repository.The base URL is https://api.picassoia.com/v1, and authentication is a Bearer token. This request asks for a five second clip:
curl -s -X POST https://api.picassoia.com/v1/models/picassoia/picassoia-video/predictions \
-H "Authorization: Bearer $PICASSOIA_TOKEN" \
-H "Content-Type: application/json" \
-d '{"input": {"prompt": "a paper boat sailing down a rainy street", "duration": 5}}'
The response returns a prediction ID. Only prompt is required, and it accepts 1 to 4,000 characters. If the call is rejected, check the credential and the plan first: an account without an Infinite plan cannot create predictions, although it can still read, list and cancel the ones it already has.
Send the ID to the predictions endpoint until the status settles:
curl -s https://api.picassoia.com/v1/predictions/PREDICTION_ID \
-H "Authorization: Bearer $PICASSOIA_TOKEN"
The status moves through starting and processing to succeeded, failed or canceled. A failure is final, so submit a fresh prediction instead of waiting. Poll every few seconds and keep your total in flight at or below the cap of 5.
| Field | What it does | Notes |
|---|---|---|
prompt | Describes the clip | Required, 1 to 4,000 characters |
image | Sets the opening frame | HTTPS URL or data URL, turns the call into image to video |
resolution | Output size | 480p default, 720p, 1080p |
duration | Length in seconds | Up to 20 at 480p, 10 at 720p, 5 at 1080p |
aspect_ratio | Frame shape | 16:9, 9:16, 1:1, 4:3, 3:4, 3:2, 2:3 |
save_audio | Keeps the audio track | Defaults to true |
seed | Repeats a result | Set it when you want a reproducible draft |
Write prompts as a short chronological scene: the subject and its starting pose, the motion over the clip, the camera move, then the light. "A woman in a cream sweater lifts a coffee mug, slow dolly in, warm morning light from a window on the left" beats "nice coffee video" every time.
💡 Tip: draft at 480p with a fixed seed, then regenerate only the winners at 720p. You keep the composition and spend the extra time only where it pays.
Prototype or side project. Use open weights if you already own a capable GPU. If you do not, use the PicassoIA API while predictions are free and keep the wrapper thin so a pricing change does not hurt.
Client work and shipped apps. Pick a provider with a published rate card and stable terms, such as Veo 3.1 Fast, and budget with the attempts formula. Get the licensing answers in writing.
High volume pipelines. Queue everything, respect the concurrency cap, draft low and finish high, and log the cost per usable clip. That single metric tells you when to switch models.
| Situation | Best fit | Why |
|---|---|---|
| Testing prompts and ideas | Open weights or the PicassoIA API | No per clip charge to worry about |
| Client deliverable | A provider with a published rate card | Stable terms and predictable cost |
| Hundreds of clips a week | A mix of low rate drafts and a premium finish | Keeps the attempts term small |
You do not need to settle the pricing question before you start. Write one prompt, run it through two models, and compare the results side by side. Open PicassoIA Video and Seedance 2.5 Lite on Picasso IA, paste the same scene description into each, and see which one gives you a clip worth keeping.
When the output looks right, copy the cURL request from this article, swap in your own prompt, and wire it into your app. Create your first video on Picasso IA today and let your own results decide which option earns a place in your stack.
