Google is shutting down three model IDs on October 22, 2026: veo-3.1-generate-preview, veo-3.1-fast-generate-preview, and veo-3.1-lite-generate-preview. (The older gemini-omni-flash-preview endpoint was already deprecated on September 30, 2026 — if your automation still calls it, it’s broken now, not on October 22.) If your Make scenario, n8n workflow, or Zapier automation calls the Gemini API with one of those IDs typed into an HTTP request field, your video pipeline goes dark on that date — no warning email, no graceful fallback, just failed API calls and broken automations.
This isn’t a rumor or a vague “sometime in Q4” announcement. Google’s own Gemini API deprecations page lists October 22, 2026 as the shutdown date for these preview models, with gemini-omni-1.1-flash (Gemini Omni Flash) named as the replacement. Multiple third-party trackers that cross-check Google’s notices confirmed the same dates in mid-October — including the Faceloom deprecation table, which is the clearest side-by-side of what’s retiring and what’s replacing it. The developer-focused migration guides are already out, but they’re written for people who read API changelogs for fun. This guide is for everyone else: the no-code builder whose reels pipeline runs on a Make scenario they set up six months ago and haven’t touched since.
Here’s the full picture: what dies, what survives, what it costs, and a concrete migration plan you can finish in an afternoon.
Who’s Actually Affected (Probably You)
The people who get hurt by this shutdown aren’t Google Cloud architects. They’re creators and small teams who followed a YouTube tutorial or a blog post that said “paste this model ID into your HTTP module” — and it worked, so they never thought about it again.
You’re affected if your setup uses any of these with the Gemini API (generativelanguage.googleapis.com):
- Make: a scenario with an “HTTP — Make a request” module hitting the Gemini API with a model ID in the URL or JSON body. Many Veo 3.1 tutorials had you paste
veo-3.1-generate-previewstraight into the request. - n8n: a workflow with an HTTP Request node calling the Gemini API, with the model name hardcoded in the URL path or in a “Set” / “Code” node feeding into it.
- Zapier: Webhooks by Zapier POSTing to the Gemini API with the model ID in the payload.
There’s one group that gets a reprieve: if your setup runs on Vertex AI (Google Cloud’s enterprise route) with the production -001 model IDs, those have a separate retirement notice — no earlier than November 17, 2026. More on that distinction later, because mixing up the two routes is the most common mistake in this whole migration.
What Exactly Dies on October 22, 2026
Let’s be precise about which model IDs are retiring, because half the confusion online comes from people conflating different Google surfaces.
| Model ID | Route | Earliest shutdown | Replacement |
|---|---|---|---|
veo-3.1-generate-preview |
Gemini API | October 22, 2026 | gemini-omni-1.1-flash |
veo-3.1-fast-generate-preview |
Gemini API | October 22, 2026 | gemini-omni-1.1-flash |
veo-3.1-lite-generate-preview |
Gemini API | October 22, 2026 | gemini-omni-1.1-flash |
One row deliberately left out: gemini-omni-flash-preview — the older preview of Omni Flash itself — was already deprecated on September 30, 2026, per Google’s August 27 release notes. Tutorials that reference it are still circulating, but its deadline has passed; if it’s in your automation, replace it today, not October 22.
A few things worth noting from Google’s own deprecation wording, as documented by third-party trackers that monitor the official page:
- These are the earliest possible shutdown dates. In practice the models could technically stay up a bit longer, but treat October 22 as the hard deadline. Planning around “maybe it’ll work until November” is how pipelines die at the worst moment.
- After shutdown, the endpoint stops working entirely. This isn’t throttling — calls fail.
- The replacement has been generally available since August 27, 2026 on the Gemini API paid tier. You’re not migrating to a beta.
One more thing that trips people up: the Gemini app and Google Flow are separate consumer surfaces. If you generate videos by hand inside Google Flow, this API notice doesn’t change your Flow credit burn rates — Flow has its own pricing. This shutdown only hits the API route, which is exactly the route your automations use.
What Survives: Vertex vs. AI Studio Is the Key Distinction
Here is where no-code builders lose the most time. Google serves video models through three different doors, and the doors have different retirement schedules.
Gemini API (AI Studio keys): dead October 22
This is generativelanguage.googleapis.com — the endpoint behind any API key you generated in Google AI Studio. The three veo-3.1-*-generate-preview IDs on this route shut down October 22, 2026. This is also the route most Make/n8n/Zapier tutorials use, because AI Studio keys are free to start with and easy to paste into an HTTP module.
Google AI Studio (the web playground): retired Veo 3.1 on October 15
Per reporting from AI video trackers in mid-October 2026, the AI Studio web interface itself retired Veo 3.1 on October 15, 2026, and Gemini Omni Flash is now its default Google video model. So if you’ve been testing prompts in the Studio playground to decide what to put in your automation, the model behind that interface has already changed out from under you. Don’t compare your old playground outputs to new ones and panic — you’re looking at two different models.
Vertex AI (Google Cloud): production models survive until November 17, 2026
Google Cloud’s Vertex AI route uses other model IDs — the -001 versions (veo-3.1-generate-001 and the Fast/Lite equivalents), GA since November 17, 2025. Google’s Vertex documentation lists their retirement date as November 17, 2026 or later — nearly four extra weeks of runway.
The migration trap here is obvious: you read “Veo 3.1 survives until November 17” on a Vertex pricing page and assume your automation is safe, while your actual calls go through the Gemini API route that dies October 22. The model ID in your workflow tells you which route you’re on. If your ID ends in -preview, you’re on the Gemini API route and October 22 is your deadline. If it ends in -001, you’re on Vertex and you have until mid-November — but you’ll still need to migrate eventually.
Meet the Replacement: gemini-omni-1.1-flash
Google’s named successor is gemini-omni-1.1-flash, part of the Gemini Omni model family that Google has been positioning as its next-generation video model since its developer preview in mid-2026. If you’ve seen “Gemini Omni” in the Gemini app or YouTube Create, this is the same model family — the API version of it.
The specs that matter for your automations
- Clip length: 3 to 10 seconds per generation, in whole seconds (Veo 3.1 did 4, 6, or 8 — so your automation’s duration parameters may need adjusting).
- Resolution: 360p, 720p, 1080p, and 4K, in 16:9 or 9:16. Google’s API docs label the 1080p and 4K options as “upscaled” — worth knowing if you promise clients “true 4K.”
- Audio: Omni generates native audio along with the video — the current docs give no parameter to switch it off. A major behavioral difference from Veo 3.1 (details below).
- Inputs: text, image, and video all supported, including conversational editing of existing clips.
- Availability: generally available on the Gemini API paid tier since August 27, 2026. No free tier for API video generation.
The price: roughly $0.10 per second at 720p
Here’s the billing change that surprises people. Veo 3.1 on the Gemini API billed per second of output:
- Veo 3.1 Standard: $0.40/sec (720p and 1080p), $0.60/sec (4K)
- Veo 3.1 Fast: $0.10/sec (720p)
- Veo 3.1 Lite: $0.05/sec (720p)
Gemini Omni Flash bills by output token, not by second: $17.50 per 1 million video output tokens, with a second of 720p video consuming roughly 5,792 tokens. Google’s own arithmetic puts that at about $0.10 per second at 720p on standard pricing — the same as Veo 3.1 Fast, and roughly a quarter of Veo 3.1 Standard. Input (text, image, video, or audio prompts) bills separately at $1.50 per 1M tokens.
What this means in practice: if your automation ran on Veo 3.1 Standard at $0.40/sec, migrating to Omni Flash cuts your per-second cost by roughly 75%. If you were on Fast, it’s a wash. If you were on Lite at $0.05/sec, your cost per second roughly doubles — though Omni’s 360p draft tier (around $0.03/sec) can offset that for iteration.
A quick reality check on a 10-second 720p clip: about $1.00 on Omni Flash versus $4.00 on Veo 3.1 Standard. At 50 clips a month, that’s $50 versus $200. The migration pays for itself in one billing cycle.
Migration Trap #1: Don’t Just Swap the ID — Re-test Everything
This is the trap that catches everyone who treats the migration as a find-and-replace job. Google’s replacement recommendation is a model pointer, not a promise of identical output. Third-party coverage of this shutdown keeps repeating the same warning: output style, clip length limits, and audio behavior can differ between models, so a straight swap is not guaranteed to look the same.
The audio problem
This is the single biggest behavioral difference. On Veo 3.1, audio generation was part of the package too — but with Omni Flash the current docs give no parameter to disable audio generation, so in practice clips arrive with a soundtrack. A developer who documented a production migration test in October 2026 found exactly this: Omni generated audio with the video, and their workflow — which layered separately-produced music over the clip — needed to mute or replace the generated track in post.
If your automation produces reels where you add your own voiceover, music, or sound design in a later step (a very common n8n pattern: generate clip → download → merge with audio in FFmpeg), your new Omni-generated clips can arrive with baked-in audio you didn’t ask for. Plan on a strip-the-audio step in your pipeline, or the doubled-up soundtrack will sound like a mess.
The style and clip-length problem
Omni Flash and Veo 3.1 render differently. Same prompt, different look — different lighting interpretations, different motion behavior, different character consistency. If your brand, your client’s brand, or your audience’s expectations are built on how Veo 3.1 renders, swapping IDs without looking will produce clips that feel “off” without being technically broken.
The practical fix is boring and essential: generate 5–10 test clips on Omni Flash with your actual production prompts before you migrate the workflow. Keep a before-and-after comparison folder. If your prompts produce noticeably worse results on Omni, you’ll need to adjust them — Omni’s conversational editing and different training mean prompt phrasing that worked on Veo may need tuning.
The parameter problem
Your automation almost certainly passes parameters like duration, resolution, and aspect ratio. Omni Flash’s valid values differ slightly: 3–10 seconds (Veo was 4, 6, 8), and its 1080p/4K outputs are upscaled rather than native. If your workflow hardcodes “8 seconds, 1080p” and Omni expects whole-second durations in its own range, a hardcoded 6-second Veo duration still works — but any automation that requests durations or resolutions outside Omni’s supported set will fail at the API level. Audit the parameters, not just the model ID.
Migration Trap #2: The Vertex vs. API Date Mix-up
The second trap is date confusion between the two routes — and it’s understandable, because even careful people mix them up.
Your migration deadline depends entirely on the route your workflow uses:
- Check the endpoint URL in your HTTP module. If it contains
generativelanguage.googleapis.com, you’re on the Gemini API route — deadline: October 22, 2026. - Check the model ID. IDs ending in
-previeware Gemini API preview models — deadline: October 22, 2026. IDs ending in-001are Vertex AI production models — retirement no earlier than November 17, 2026. - Check which API key you’re using. A key generated in Google AI Studio means the Gemini API route. A key tied to a Google Cloud project with Vertex AI enabled means the Vertex route.
The dangerous middle ground: some setups use a Google Cloud project key but call the generativelanguage.googleapis.com endpoint. That combination is on the Gemini API route regardless of where the key came from. The endpoint and model ID determine your deadline — not the key’s origin.
If you’re on Vertex with -001 models, you have until November 17 at the earliest — but don’t treat that as leisure time. Vertex’s eventual migration will presumably point at Omni models on the Vertex route too, and the audio/style re-testing discipline applies there just the same.
How to Find Every Hardcoded Model ID in Your Automations
This is where migrations actually fail — not on the new model, but on the old reference you forgot existed in a scenario you cloned last spring.
Make: export the blueprint and search it
Open your scenario → three dots → Export blueprint (a JSON download). Search it for veo-3.1 and gemini-omni-flash-preview. You’ll find hits in HTTP module URLs (like .../v1beta/models/veo-3.1-generate-preview:predictLongRunning) and JSON request bodies. Repeat for every video scenario — cloned scenarios are the classic miss.
n8n: check HTTP Request nodes and Code nodes
Check every HTTP Request node for the Gemini API URL — the model ID sits in the path. Then check Code and Set nodes, since many workflows build the URL dynamically from a variable. Export the workflow as JSON and grep it the same way as a Make blueprint.
Zapier: inspect Webhooks steps
Open each relevant Zap, expand Webhooks by Zapier steps, and check the URL and custom payload fields. Zapier lacks a clean bulk export, so this is a manual pass — including dormant Zaps you might re-enable later, since a dead model ID in a sleeping Zap is a failure waiting to happen.
Audit checklist (before you change anything)
- [ ] List every automation that generates AI video — active and dormant.
- [ ] For each, record: platform, model ID, endpoint URL, duration/resolution parameters, and whether you add your own audio downstream.
- [ ] Flag pipelines that layer separate audio over clips — those need an audio-strip step after migration.
Do the audit before you touch a single module. Changing IDs first and discovering a forgotten scenario later leaves you with half your pipeline on one model and half on another.
The Migration, Step by Step
Budget an afternoon.
Step 1: Confirm your key tier. Omni Flash requires the Gemini API paid tier — no free tier. If your setup ran on a free-tier AI Studio key, you’ll need a paid-tier key before anything works.
Step 2: Generate test clips. Run 5–10 generations on gemini-omni-1.1-flash with your actual production prompts. Compare against your Veo 3.1 outputs: visual style, audio behavior, and whether your duration/resolution combos are supported.
Step 3: Adjust prompts if needed. If Omni’s output looks off with Veo-tuned prompts, revise them. Omni’s conversational editing lets you iterate in the AI Studio playground with plain-language follow-ups before locking prompts into your automation.
Step 4: Update the model ID in each automation. Replace the -preview ID with gemini-omni-1.1-flash in the URL or payload. Check for duplicates — some workflows pass the model ID in two places (the request URL and a logging step).
Step 5: Add audio handling if you layer your own sound. If your pipeline merges clips with separate music or voiceover, add a step that strips the generated audio track (FFmpeg’s -an flag is the standard tool) before your merge step.
Step 6: Update your cost math. Budget roughly $0.10/sec at 720p on Omni Flash versus your old Veo tier. And fix any cost-logging formulas: Omni bills tokens ($17.50 per million output video tokens, 5,792 tokens per 720p second), so an old seconds × Veo-rate formula will lie to you quietly for months.
Step 7: Run a full end-to-end test. Trigger each migrated automation manually. One documented quirk: the Gemini video API’s download URIs require your API key on the download request — fetching without the key header returns a 403 that looks like a generation failure.
Step 8: Calendar a reminder for November 17. Even after this migration, mark the Vertex retirement date. Google rotates models fast; checking deprecation notices quarterly will save you from this scramble again.
What It Costs: The Honest Math
Here’s the per-clip math at the most common settings, using Google’s published rates:
| Setup | 8-second clip, 720p | 8-second clip, 1080p |
|---|---|---|
| Veo 3.1 Standard (old) | $3.20 | $3.20 |
| Veo 3.1 Fast (old) | $0.80 | $0.96 |
| Veo 3.1 Lite (old) | $0.40 | $0.64 |
| Gemini Omni Flash 1.1 (new) | ~$0.80 | ~$1.20 |
Two observations. First, if you were on Standard, the migration is a massive cost cut — your 8-second clips drop from $3.20 to about $0.80. Second, if you were on Lite, per-second cost goes up, but you get a meaningfully more capable model, and the 360p draft tier (~$0.03/sec) covers cheap iteration.
One more cost note: input tokens bill at $1.50 per million. Large reference images or video inputs add a small amount per generation — minor next to output costs, but a new line item your old Veo budgets didn’t have.
When You Can Hold Off
Not everyone needs to move on October 22:
- You’re on the Vertex AI route with
-001models. You have until November 17, 2026 at the earliest. Use the extra weeks to test properly rather than rushing. - You only generate in Google Flow by hand. Flow is a separate consumer product with its own credit system — the API shutdown doesn’t touch your Flow generations.
- Your pipeline is dormant. Just document the dead model ID somewhere so future-you doesn’t waste an afternoon debugging a failed call in six months.
What you should not do: keep generating on the preview IDs past October 22 hoping for a grace period. Google’s deprecation language says the endpoint stops working — there’s no published grace window, and “earliest possible date” cuts both ways.
A Quick Word on Alternatives
If this episode has you questioning single-vendor dependence — fair. The lesson of the 2026 shutdowns (Veo 3.1 previews on October 22, the Sora API on September 24) is that a hardcoded model ID is a dependency with a retirement date, and every vendor rotates models on roughly this cadence.
If you want to compare the field before committing to Omni Flash, our best AI video generators for beginners covers the current landscape, and the Kling 3 vs. Veo cost comparison is useful if price-per-second decides it. (Our Veo 3.1 reels usage guide covers the creative side — but its API references point at the retiring preview IDs, so don’t copy model IDs from it into new automations.)
The pragmatic middle ground: migrate to Omni Flash now (it’s Google’s stated default), but keep your model ID in a single variable per automation rather than hardcoded in three places. Next time a model retires — and there will be a next time — you’ll change one field instead of hunting through JSON exports.
Frequently Asked Questions
Will my Make/n8n/Zapier automation break on October 22, 2026?
If it calls the Gemini API with veo-3.1-generate-preview, veo-3.1-fast-generate-preview, or veo-3.1-lite-generate-preview, yes — those endpoints are scheduled for shutdown on that date, and Google says shut-down endpoints stop working entirely. (The older gemini-omni-flash-preview was already deprecated on September 30, 2026 — if your automation calls it, it’s broken now.) If your automation uses Vertex AI -001 model IDs instead, it keeps running until at least November 17, 2026.
What should I replace the Veo 3.1 preview model ID with?
Google names gemini-omni-1.1-flash (Gemini Omni Flash) as the replacement for the three retiring Veo 3.1 preview models. It’s been generally available on the Gemini API paid tier since August 27, 2026, costs roughly $0.10 per second at 720p, and supports 3–10 second clips at up to 4K (upscaled), generating native audio with each clip.
Is Gemini Omni Flash cheaper than Veo 3.1?
Compared to Veo 3.1 Standard ($0.40/sec), yes — roughly a quarter of the price. Compared to Veo 3.1 Fast ($0.10/sec), it’s about the same. Compared to Veo 3.1 Lite ($0.05/sec), it’s roughly double per second, though the 360p draft tier (~$0.03/sec) covers cheap iteration. Note the billing unit changed from per-second to per-token ($17.50 per million output video tokens).
Why can’t I just swap the model ID and keep my prompts?
You can swap the ID, but you shouldn’t skip testing. Omni Flash generates native audio as part of its clips (audio references are unsupported and voice editing is restricted in the current docs), renders in a different style than Veo 3.1, and accepts slightly different duration values (3–10 seconds vs. Veo’s 4/6/8). If your pipeline adds its own music or voiceover, plan on a step that strips the generated audio track. A developer who documented a production migration found exactly this: their workflow needed a mute step after switching.
Does this affect videos I make in Google Flow or the Gemini app?
No. This shutdown targets the Gemini API route used by automations and code. Google Flow and the Gemini app are separate consumer surfaces with their own model availability and credit systems. (The AI Studio web playground did retire Veo 3.1 on October 15, 2026 in favor of Omni Flash, but that’s the testing interface, not your automations.)
How do I find every place the old model ID is used?
Export your automation’s full definition and search it: Make has “Export blueprint” (downloads JSON you can search for veo-3.1), n8n workflows can be downloaded as JSON and grepped the same way, and Zapier requires a manual pass through Webhooks steps since it lacks a clean bulk export. Check dormant and cloned automations too — the copy you forgot about is the one that breaks at 2 AM.