Veo 3.1 After Preview Shutdown: IDs and Audio Costs

A Preview shutdown is not a shutdown of every Veo 3.1 model

If an old Veo tutorial suddenly fails, inspect the complete model identifier before rewriting the creative prompt. Google's current Gemini API documentation says the Veo 3.1 Preview identifiers were shut down on October 22, 2026. The same notice points readers toward Gemini Omni Flash or GA models on its cloud platform. That is a specific product and identifier change, not evidence that every service using the words Veo 3.1 has stopped.

On November 1, Orux's public catalog lists veo-3.1-generate-001 and veo-3.1-fast-generate-001. This guide separates those listed aliases from the retired Gemini API Preview names, checks the controls actually exposed by Orux, and calculates how switching generated sound changes a short-clip budget. It delivers a runnable local calculator and its observed output. It does not claim that a catalog entry proves a successful video generation, nor compare the visual quality or speed of these models.

For a producer planning silent rehearsal footage under an independently edited soundtrack, the most useful finding is concrete: the checked Orux rows price silent and sound-enabled generation differently, while 720p and 1080p have the same rate within each model/audio combination. Turning off sound affects this snapshot's calculation; merely selecting 720p does not. Decide the sound requirement before reserving a batch of shots.

Keep the Preview name, GA name and service boundary separate

The retired Gemini API names include veo-3.1-generate-preview, veo-3.1-fast-generate-preview and veo-3.1-lite-generate-preview. Do not shorten these to “Veo” in a migration checklist. A suffix can determine whether an example addresses a retired Preview or a different published identifier. Google's separate cloud model page documents Veo 3.1 GA identifiers; it is a separate reference, not permission to reuse every cloud request field on Orux.

For Orux, copy the exact alias from the model page and check that page's current form. The two aliases examined here end in -001; neither is the retired -preview alias. This establishes a naming distinction and current catalog availability. It does not establish account access, present generation health, content-policy acceptance or an identical response to an old prompt. Those require a deliberately authorized request and inspection of its actual result.

When reviewing an existing project, record the service URL, full model alias, date of the reference example and options used. An old sample can be useful for scene construction while its endpoint or identifier has become obsolete. Preserve the scene brief separately so that changing a technical identifier does not silently change the creative task you intended to evaluate.

Use the controls that Orux actually lists

Both checked Orux forms expose a prompt, an eight-second duration option, 720p or 1080p resolution, an aspect-ratio selector and a boolean generate_audio. Eight seconds is the only duration value listed in these particular forms. Do not replace it with six just because another provider's documentation or a different model supports a shorter clip. The calculator below rejects that unchecked assumption rather than producing a plausible but irrelevant quote.

The forms also list 16:9, 9:16 and 1:1 ratios. A listed choice is a configured UI option, not a benchmark of how well every composition will work. Frame the subject for the intended destination and inspect the returned file dimensions after an actual request. Similarly, a boolean for sound does not prove that a requested line of speech, musical rhythm or precise event synchronization will be delivered correctly.

The checked forms do not expose image-reference or video-extension fields. This observation describes these forms on this date; it is not a claim that Veo as a model family lacks those capabilities. If your task depends on preserving a particular photographed location, do not treat a text-only scene suggestion as proof that the generated footage will document the real place. Verify a suitable input path before paying for that task.

Compare sound choices in Credits before choosing a batch

Swipe horizontally to view the full table.

Prices are the public Orux per-second rows checked on November 1, 2026; one Credit equals $0.005. Each displayed row has exact quantity rounding and no additional minimum charge in the checked configuration. Match both resolution and generate_audio to the selected row. These figures are calculations from configured rates, not a receipt for generated footage and not an official-provider savings comparison.

For an eight-second Fast clip, sound adds 89.0016 Credits. For the other listed alias, it adds 355.9992 Credits. The numbers are intentionally left at the catalog's precision: rounding them first and multiplying later can obscure small differences across a larger batch. Neither the larger rate nor the word Fast establishes which output a particular editor would keep.

Run the calculator without making a video request

Save the following as budget.py and run python3 budget.py. Python's standard-library Decimal performs the displayed arithmetic from explicit decimal strings. The script reads no API key, opens no network connection and spends no Orux Credits. It is a dated offline snapshot, so update the rates and allowed options from the live model pages before using it for a future approval.

"""Offline Orux price snapshot, 2026-11-01. No generation request."""
from decimal import Decimal
import json

RATES = {
    "veo-3.1-fast-generate-001": {False: "22.2501", True: "33.3753"},
    "veo-3.1-generate-001": {False: "44.5002", True: "89.0001"},
}

def quote(model, audio, seconds=8, resolution="720p"):
    if model not in RATES or type(audio) is not bool:
        raise ValueError("Use an exact listed alias and a boolean audio flag")
    if seconds != 8 or resolution not in {"720p", "1080p"}:
        raise ValueError("Outside the checked Orux form options")
    return Decimal(RATES[model][audio]) * seconds

if __name__ == "__main__":
    report = {}
    for model in RATES:
        silent, sound = quote(model, False), quote(model, True)
        report[model] = {"silent_8s_credits": str(silent),
                         "sound_8s_credits": str(sound),
                         "audio_extra_8s_credits": str(sound-silent),
                         "four_sound_clips_credits": str(sound*4)}
        assert quote(model, True, resolution="1080p") == sound
    for bad in [("veo-3.1-generate-preview", False, 8),
                ("veo-3.1-generate-001", "false", 8),
                ("veo-3.1-generate-001", False, 6)]:
        try:
            quote(*bad)
        except ValueError:
            pass
        else:
            raise AssertionError("Invalid assumption accepted")
    print(json.dumps(report, indent=2))

The actual local output was:

{
  "veo-3.1-fast-generate-001": {
    "silent_8s_credits": "178.0008",
    "sound_8s_credits": "267.0024",
    "audio_extra_8s_credits": "89.0016",
    "four_sound_clips_credits": "1068.0096"
  },
  "veo-3.1-generate-001": {
    "silent_8s_credits": "356.0016",
    "sound_8s_credits": "712.0008",
    "audio_extra_8s_credits": "355.9992",
    "four_sound_clips_credits": "2848.0032"
  }
}

The checks reject a retired Preview name, a string containing the word false instead of a real boolean, and an unlisted six-second duration. They also verify that the displayed 1080p sound quote equals its 720p counterpart for both aliases. These tests validate the local quoting rules; they do not test a remote model, complete a migration or establish that a paid request will succeed.

The boolean check is worth retaining when adapting this small tool. In many programs, a nonempty string such as "false" is truthy. Treating it as an audio choice could select a different cost row from the one the producer intended. Reject ambiguous input and make the person approving a batch select an explicit option rather than guessing on their behalf.

Plan a 30-second edit from four eight-second sources

Consider a proposed dance teaser: a dancer entering, a turning-shoe detail, a wide stage view and a closing composition with space for an event title. Four source clips of eight seconds provide 32 seconds before editing. An editorial plan of 8 + 8 + 8 + 6 seconds produces a 30-second cut by trimming two seconds from the final source. This is an edit plan, not footage generated for this article.

Generation is budgeted for the full 32 source seconds, not just the 30 seconds kept in the edit. Four silent Fast clips calculate to 712.0032 Credits; four sound-enabled Fast clips calculate to 1068.0096 Credits. With veo-3.1-generate-001, the corresponding totals are 1424.0064 and 2848.0032 Credits. Additional attempts and separately sourced music or editing are outside these totals.

If the teaser will receive one continuous licensed soundtrack in post-production, decide whether separate generated sound for each source contributes anything you intend to use. If footsteps or stage movement is central, sound-enabled generation becomes a candidate worth evaluating. This is a task-based choice, not a claim that turning sound off always creates a better result or guarantees a lower final production cost.

Approve the next experiment as a separate decision

Before spending on generation, retain the exact alias, sound choice, resolution, source duration, number of planned attempts and maximum authorized cost. Use a small, clearly described scene to evaluate the relevant decision. Review the resulting file and sound together before scaling the plan. A successful HTTP response or a listed model is not a substitute for inspecting the delivered media.

The earlier Seedance shot-budget guide handles a different pricing setup; its arithmetic should not be copied into these per-second rows. The Kling shot comparison explains a separate editorial task: choosing between actual videos. Here the contribution is separating a recent Preview retirement from today's listed aliases and making the audio decision visible before a generation budget is approved.

Prices checked on 2026-11-01. Sources and current model pages: Google video guide, Google GA reference, Public catalog, Public catalog. veo-3.1-fast-generate-001, veo-3.1-generate-001.

FAQ

Q.Did Google shut down every Veo 3.1 service?

No. The cited notice names the Gemini API Preview identifiers and their October 22, 2026 shutdown. It also points to other options, including GA models. Check the complete identifier and service boundary. Orux catalog availability and a successful paid generation are separate facts; this article verifies the former, not the latter.

Q.Can I copy a six-second setting from another Veo tutorial?

Not on the evidence checked here. These Orux forms list only eight seconds, and the local calculator rejects six. A broader vendor capability or another service's option does not prove the same control is available in this form. Recheck the current model page before authorizing a request that depends on a different duration.

Q.Does choosing 720p reduce the displayed generation price?

No, not in these checked rows. The 720p and 1080p entries match within each model and sound setting. That does not prove equal visual quality, delivery speed or suitability. Choose resolution for the deliverable, then match the exact current price conditions rather than assuming a smaller label means a lower charge.

Q.Are the four-shot totals a guaranteed final production budget?

No. They cover one planned set of four eight-second source clips at the dated configured rates. They exclude extra attempts, editing and separately sourced audio. Trimming two source seconds does not reduce the number of generated seconds used in this plan. Inspect the actual estimate and set a bounded experiment budget before generation.

Q.Did this article generate or compare Veo videos?

No. The observed output is from the offline price calculator, and the rehearsal sequence is a proposed edit plan. There is no visual-quality ranking, latency measurement or generated sound evaluation here. Use the verified alias and cost conditions to design a separate, authorized media test if that evidence is needed.