Veo 3.1 API Name: Model IDs for Veo 3.1, Fast and Lite

You paste a model name into your script, run it, and get an error instead of a video. With Veo 3.1 that happens more often than it should, because the name in a launch post, the name in the Gemini API and the name on a third-party platform are three different strings. This page lists the exact IDs for Veo 3.1, Veo 3.1 Fast and Veo 3.1 Lite, what each one can do, what each one costs per second, and how to call them without guessing.

The short version: in the Gemini API the three names are veo-3.1-generate-preview, veo-3.1-fast-generate-preview and veo-3.1-lite-generate-preview. Everything below explains why they look like that, which one fits your job, and how to run the same models on PicassoIA when you would rather click than code. If a request fails right away, compare your string against the table below before you touch anything else, because a one-character slip is easy to make and hard to spot in a long script.

The Three Model IDs

Google's Gemini API documentation lists these codes for the Veo 3.1 family. Copy them exactly, including the -preview ending.

💡 Tip: IDs are literal strings. veo-3.1, veo3.1 and veo-3-1-fast are not any of the three codes above, so a request that uses them is asking for a model that Google's list does not contain.

Veo 3.1 standard

veo-3.1-generate-preview sits at the top of the family. Per Google's property table it takes text or an image as input, returns video with natively generated audio, and offers 720p, 1080p and 4k output. Clips run 4, 6 or 8 seconds, and 1080p and 4k are limited to 8 seconds. Pick it when the clip is the final deliverable and the budget allows it.

Veo 3.1 Fast

veo-3.1-fast-generate-preview shares the same input, output, resolution and duration rows as the standard model in Google's table. What changes is the trade: a lower price per second and shorter waits. In the example generations published on its PicassoIA page, Fast clips finished in roughly 45 to 59 seconds, against 72 to 114 seconds for the examples on the standard Veo 3.1 page.

Veo 3.1 Lite

veo-3.1-lite-generate-preview is the cheapest tier. It supports 720p and 1080p (8 seconds only), has no 4k output, and Google's documentation says video extension is not available for it. The examples on the Veo 3.1 Lite page finished in about 37 to 42 seconds. Think of it as the volume tier: lots of drafts, social clips and background loops.

Where the Names Come From

Reading the pattern

Every ID follows one shape: version, optional tier, the word generate, then a release stage. So veo-3.1-fast-generate-preview reads as Veo 3.1, Fast tier, generate, preview stage.

The older Veo 3 models use a different stage suffix. veo-3.0-generate-001 and veo-3.0-fast-generate-001 are listed as stable, with 8-second clips at 720p or 1080p. If you want those on PicassoIA instead, they are Veo 3 and Veo 3 Fast.

Preview means Google can still change behavior, limits or price. Put the string in one config constant so a single edit moves your whole project the day a name changes.

Stable or preview? If your product cannot tolerate a model being renamed or repriced, the stable Veo 3 pair is the safer base, as long as 8-second clips at 720p or 1080p are enough. If you need 4, 6 or 8 second lengths, the cheaper Lite tier or 4k output, you are in preview territory and should plan for change. Log the model string next to every generated file so you can reproduce or audit a result months later.

Names on other platforms

Never paste a Gemini API string into another platform. PicassoIA catalogs these models under the owner google, with names that drop the generate and preview words, and page slugs that drop the dots.

The practical rule: a tool that talks to the Gemini API wants veo-3.1-generate-preview, while a tool or catalog that works with owner and name pairs wants google/veo-3.1. Mixing the two is an easy way to end up asking for a model that does not exist.

Google Cloud's Vertex AI keeps its own model listing. Read the ID from that console instead of assuming it matches the Gemini API string.

Specs Side by Side

All numbers below come from Google's Gemini API documentation for Veo.

The 1080p rule

Google ties 1080p (and 4k on the models that have it) to 8-second clips, and PicassoIA's Lite form repeats the same rule: duration must be 8 when you choose 1080p. Treat resolution and duration as a pair. A 4 or 6 second clip means 720p.

What Lite leaves out

Beyond 4k and video extension, the Lite form on PicassoIA has no negative prompt field, no audio toggle and no reference images. The standard model's form has all three, and Fast keeps the negative prompt and the audio toggle. If your pipeline depends on reference images to hold a character steady, Lite is the wrong tier.

What Each Call Costs

Google bills Veo by the second of generated video. These are the with-audio prices on its pricing page when this was written, plus the cost of one 8-second clip.

💡 Check before you budget: these are preview models, and preview prices can move. Google also states that you are only charged when a video is successfully generated.

Draft cheap, finish expensive

Run the numbers on a real batch. Ten 8-second drafts on Lite at 720p cost 10 x $0.40 = $4.00. Three refinements on Fast at 1080p cost 3 x $0.96 = $2.88. One final on standard at 1080p costs $3.20. Total: $10.08. Making all fourteen clips on standard 1080p would cost 14 x $3.20 = $44.80 for the same number of generations.

Shorter drafts save even more. A 4-second clip is allowed at 720p, so a Lite draft costs 4 x $0.05 = $0.20, a Fast draft costs $0.40 and a standard draft costs $1.60. Settle the camera move and the sound on 4-second drafts, then spend the 8 seconds once the prompt works.

On PicassoIA you generate from the browser, so there is no per-second meter in your own code to track. Limits depend on your plan, so check it before a large run.

A Working Python Call

Google's documentation shows the pattern below for the standard model. It uses the google-genai SDK, starts a long-running operation and polls it until the video is ready. The client reads your Gemini credential from the environment, so set that up the way Google AI Studio describes before you run it.

import time
from google import genai
from google.genai import types

client = genai.Client()

prompt = "A slow dolly shot along a rainy street at dusk, a street musician plays saxophone, rain and distant traffic in the audio"

operation = client.models.generate_videos(
    model="veo-3.1-generate-preview",
    prompt=prompt,
)

while not operation.done:
    print("Waiting for video generation to finish...")
    time.sleep(10)
    operation = client.operations.get(operation)

generated_video = operation.response.generated_videos[0]
client.files.download(file=generated_video.video, destination="output.mp4")

Copy Prompt

Moving between tiers changes one argument. Keep the three names in a dictionary and pass the one you need:

VEO_MODELS = {
    "standard": "veo-3.1-generate-preview",
    "fast": "veo-3.1-fast-generate-preview",
    "lite": "veo-3.1-lite-generate-preview",
}

operation = client.models.generate_videos(
    model=VEO_MODELS["fast"],
    prompt=prompt,
)

Copy Prompt

Polling the operation

The call returns an operation, not a video. Your loop checks operation.done every 10 seconds and then downloads the file. Budget for waits of a minute or two on the standard model and less on Fast and Lite, based on the example timings above. Do not resubmit while an operation is still running, or you generate (and pay for) a second clip.

Mistakes that waste money

• Missing suffix. veo-3.1-generate without -preview is not a listed code.
• 4k on Lite. The tier does not offer it.
• 1080p with a 6 second clip. Set 8 seconds, or drop to 720p.
• Extending a Lite clip. Google says extension is not available for Lite.
• Scattering strings. Hard code the ID in ten files and a rename costs an afternoon.

Before you queue a batch, run through this short checklist:

• The model string matches one of the three codes in the first table, character for character.
• Resolution and duration are a valid pair: 1080p means 8 seconds.
• The tier you picked supports what you need: 4k, extension and reference images are not on Lite.
• You logged the model string and the prompt next to each output file.
• You multiplied the per-second price by the number of clips before pressing run.

Which Model Fits Your Job

Fast for iteration

Most prompts need three or four rewrites before the camera move and the sound land where you want them. Do those rewrites on Veo 3.1 Fast, then promote the winning prompt to the standard model.

Because every tier returns native audio, write the sound into the prompt. PicassoIA's own example prompts do exactly that: a train whistle and ambient station noise, soft jazz with rain pattering against glass, a line of spoken dialogue in quotes. A prompt that names the sound gives the model something to build the audio track around, and you will hear the difference between drafts quickly.

Standard for finals

When a clip will sit on a landing page, in a client deck or on a big screen, the extra cost of Veo 3.1 is easy to justify. It is also the only tier in this family with reference images on PicassoIA, and it shares the 4k row with Fast in Google's table.

Lite for volume

Veo 3.1 Lite fits any workflow where you need dozens of short clips: product loops, vertical posts, background footage for a podcast. Switch the aspect ratio to 9:16 and set 720p for the cheapest result.

How to Use Veo 3.1 on PicassoIA

No code, no credentials, same family of models. Here is the flow on the Veo 3.1 page:

1. Open the model page. Go to Veo 3.1 on PicassoIA.
2. Write the prompt. This is the only required field. Describe the subject, the camera move, the light and the sound you want to hear.
3. Set duration. Choose 4, 6 or 8 seconds. The default is 8.
4. Set resolution and ratio. Choose 720p or 1080p (the default is 1080p), and 16:9 or 9:16.
5. Keep audio on. The audio option is enabled by default. Add a negative prompt for anything you want left out.
6. Generate and download. Wait for the clip, then save the file.

Frames and reference images

Add an input image to start from your own first frame. Add a last frame as well and the model builds a transition between the two. Ideal images are 16:9 or 9:16 at 1280x720 or 720x1280, matching the ratio you pick.

To keep a subject consistent, upload one to three reference images instead. This works on the standard model only, and only with 16:9 and 8 seconds. If you provide reference images, the last frame is ignored, so choose one approach per clip.

Fast and Lite on PicassoIA

Veo 3.1 Fast defaults to 1080p and keeps the negative prompt and audio toggle. Veo 3.1 Lite defaults to 720p and requires 8 seconds whenever you choose 1080p.

💡 Coding against PicassoIA? When this was written, PicassoIA's developer API listed four of its own models, including PicassoIA Video and Seedance 2.5 Lite, rather than Google's Veo models. The Veo IDs in this article belong to the Gemini API, and the Veo pages above are for the browser.

Make Your First Clip Today

Pick one prompt and run it three times: once on Veo 3.1, once on Fast and once on Lite. Keep the seed the same where the form offers one, then put the three clips side by side. You will see the price and speed difference in about five minutes, and you will know which tier earns a place in your workflow.

A good test prompt: a barista slides a latte across a marble counter, steam rising, soft jazz and the hiss of an espresso machine, slow push-in on the cup. It has a clear subject, one camera move and a defined soundscape, so differences between tiers show up in motion quality and audio rather than in guesswork.

Picasso IA puts all three on one site, so you can test before you write a line of integration code. Open Veo 3.1, Veo 3.1 Fast or Veo 3.1 Lite, type the scene you have been picturing, and watch it turn into a clip with sound.