NewsAdGraam raises ₹3 crore in seed funding
AI Video & Image News

Omni LoRA Tuning: Google Video Fine-Tuning Leak

AdGraam News Desk5 min read
A glass film reel adjusted by small brass dials under studio lighting, symbolizing video model fine-tuning

The short answer

A spec called "Omni LoRA tuning" reportedly appeared in the Vertex AI API on Nov 14, 2026, for fine-tuning Google's Gemini Omni video model, according to @testingcatalog citing aitrackerbot. One account says it covers text to video and image to video. Google has not announced it, and no pricing, dates or access terms are public.

On this page (11 sections)

What is Omni LoRA tuning?

Omni LoRA tuning is a spec that reportedly appeared in the Vertex AI API for fine-tuning Google's Gemini Omni video model with LoRA. According to @testingcatalog's Nov 14 daily brief, it surfaced via aitrackerbot. Google has not announced it.

A small adapter module attaching to a large machine, representing a LoRA layer on a video modelA small adapter module attaching to a large machine, representing a LoRA layer on a video model

LoRA (low-rank adaptation) trains a small set of added weights instead of retraining a whole model. For video, that would mean teaching the model a specific product, character or visual style from your own material.

A post from @Israfilv2 says the spec covers text to video and image to video, and frames it as a threat to Higgsfield. That is the poster's opinion and reading of the spec, not a Google claim.

Omni LoRA tuning release date and availability

No release date, region list or access tier has been announced. The spec was first seen on Nov 14, 2026, when @testingcatalog included it in its daily brief. The Nov 16 brief quotes that earlier edition.

A spec appearing in an API does not mean the feature is live. Google has not said whether or when it will open to users.

Google's public tuning docs, with the LoRA page last updated Nov 12, 2026, describe parameter-efficient tuning and LoRA for LLMs on its Gemini Enterprise Agent Platform. Those pages do not mention Omni or video.

Omni LoRA tuning key features and specs

The only reported details are LoRA fine-tuning for Gemini Omni video, with text to video and image to video modes. Everything else is not announced.

SpecDetails
ModelGemini Omni (video)
MethodLoRA fine-tuning
ModesText to video, image to video (per @Israfilv2)
SurfaceVertex AI API
ResolutionNot announced
Max durationNot announced
Dataset size or formatNot announced
PricingNot announced
StatusRumor

For context, Google's LLM docs say LoRA is about 66% faster than QLoRA and up to 40% less expensive, with similar accuracy gains. QLoRA uses about 75% less peak GPU memory. Those figures apply to text models and should not be assumed for video.

How would Omni LoRA tuning work for AI video and static ads?

If it ships as described, brands could train an adapter on their own product footage or stills and generate new clips that reflect the product's look. Nothing public says how well it would hold a product's shape, packaging or logo, so that is the first thing to test once access exists.

A product photography desk with identical sneakers and a camera, representing brand-consistent ad videoA product photography desk with identical sneakers and a camera, representing brand-consistent ad video

Likely uses for D2C teams, all unconfirmed:

  • Product b-roll reused across many hook variants.
  • A fixed brand look across image to video clips built from static ads.
  • Recurring characters or mascots across a campaign.

The details that decide whether this is useful are still unknown: dataset requirements, training time, cost per run, whether adapters can be exported, and the rules on training with third-party likenesses.

A practical way to prepare is to organize the assets a tuning job would likely need. That means clean product shots from several angles, consistent lighting, and approved brand references. Even without tuning, the same library feeds image to video workflows today.

Teams can compare current video models side by side in the AdGraam Playground and chain steps in Flow Builder. For static work, see how to make static ads with AI.

Omni LoRA tuning pricing and free access

Pricing and free access are not announced. Nothing in the reports says how Omni tuning would be priced, whether training and generation would be billed separately, or whether a trial exists.

Google Cloud's docs show a "Start free" link for the platform generally. That does not indicate Omni tuning would be covered.

For a sense of how other providers describe adapter training, Together AI's changelog lists LoRA post-training for text and vision-language models, with limits such as rank up to 32 and up to three adapter sessions per resource. That is a different product and says nothing about Google's terms.

Reference pointLoRA detailMedia type
Google Omni LoRA tuningReported spec, no limits announcedVideo
Google LLM LoRA docsAbout 66% faster than QLoRAText
Together AI Nemotron 3.5 Super VL post-trainingRank up to 32, 3 adapter sessionsVision-language

Omni LoRA tuning vs other video customization options

Higgsfield is the only competitor named, by @Israfilv2, and the comparison there is rhetorical. No benchmarks or side-by-side outputs have been shared.

A fair comparison has to wait for real specs. What would decide it: training cost, time to a usable adapter, how well likeness holds, and whether you can run the result outside Google's cloud.

Tools like Higgsfield focus on a creator-facing interface for generating video. Performance teams usually need more: several models in one place, competitor ad remakes with Copy Ads, and Meta results feeding the next round of creative. A tuning option from a single model maker would cover only the first step of that loop.

Other Google video news is in BouncyBohr: Leaked Google Video Model.

What to watch next

Watch for Google documentation naming Gemini Omni tuning, a Vertex AI release note, or pricing. Also watch whether the spec lists limits on clip length, resolution and training data volume.

Until Google confirms, treat the feature as a rumor and avoid building a production plan around it.

From the timeline

What people are posting

GOOGLE IS ABOUT TO LET YOU TRAIN ITS VIDEO MODEL ON YOUR OWN STUFF 😳 its so over for higgsfield A new "Omni LoRA tuning" spec just showed up in the Vertex AI API. What's inside: • LoRA fine-tuning for Google's Omni video model • Text to video AND image to video • Train on https://t.co/XlQu9rzpb6
@Israfilv223 likesView on X
DAILY AI BRIEF 🗞 — Nov 16 ALIBABA 🔥: - Qwen4-Max is out, a 2.4T MoE model with native vision and 1M context at $1.50/$4.50 per 1M tokens. TENCENT 🔥: - Tencent launched Hy4 Flash, a 1M-context multimodal model at $0.07/$0.281 per 1M tokens, with 50% off through November 30.
@testingcatalog12 likesView on X

Key takeaways

  • An "Omni LoRA tuning" spec reportedly appeared in the Vertex AI API around Nov 14, 2026.
  • It is described as LoRA fine-tuning for Gemini Omni video, per @testingcatalog, citing aitrackerbot.
  • @Israfilv2 says it covers both text to video and image to video.
  • Google has not announced the feature, pricing, limits or release date.
  • Google's tuning docs list LoRA as a supported parameter-efficient method for LLMs, but those pages do not mention Omni video.
Run today's top models

Compare 200+ AI video and image models side by side

Run Kling, Veo, Sora, Seedance, Flux, Ideogram and Recraft in one Playground, or turn any product URL into ready-to-launch ads.

Questions, answered

Frequently asked

What is Omni LoRA tuning?

It is a spec that reportedly appeared in the Vertex AI API for fine-tuning Google's Gemini Omni video model with LoRA. The report comes from @testingcatalog, citing aitrackerbot. Google has not described it publicly.

When is Omni LoRA tuning coming out?

No release date has been announced. The spec was first reported on Nov 14, 2026, and Google has not confirmed it.

How much does Omni LoRA tuning cost?

Pricing is not announced. Nothing in the reports mentions training or inference rates.

Does it support image to video?

According to @Israfilv2, the spec covers both text to video and image to video. That is a third-party reading of the API spec, not a Google statement.

Can I use it for commercial ads?

Terms are not announced. Check Google's license and usage terms once the feature is documented.

Where this comes from

Sources

  1. LoRA and QLoRA recommendations for LLMsdocs.cloud.google.com
  2. Introduction to tuning | Gemini Enterprise Agent Platformdocs.cloud.google.com
  3. DAILY AI BRIEF 🗞 — Nov 14 OPENAI 🔥: - ChatGPT Wallet ...x.com
  4. Changelog - Together AI docsdocs.together.ai

This story is still unconfirmed: it is based on what people are posting, and the company behind it has not announced it yet. We will update this page as it develops.

AI Video & Image Newsomni lora tuninggemini omni videovertex ai video fine-tuninggoogle video model loraai video fine-tuning for adsgemini omni
Keep reading

More AI model news & creative guides

Your move

See your product as a winning ad

Turn a product URL into static ads and UGC-style videos, rebuild the competitor ads that already work, and read your Meta data to know what to make next.

AdGraam News DeskAI video and image news

The AdGraam News Desk (part of AdGraam, also searched as Adgram or Adgram AI) tracks AI image and video model launches, leaks and rumors from labs, leakers and creators, and checks each claim against its sources. Unconfirmed stories say so, and each page is updated as its story develops.

LinkedIn