Three of the free Google AI video generators tested in September 2026 were worth keeping: Vids, Flow and Pomelli. Vids covers prompt-to-video, avatars and text-based clip edits; Flow adds an agent, character elements and timestamp prompting; Pomelli starts from a URL and produces on-brand campaigns. The other tools tested did not clear that bar.
Free Google AI Video Generators: Which Three Are Worth Using
The free Google AI video generators worth using are Google Vids, Google Flow and Pomelli. Vids handles prompt-to-video, avatars and text-based clip edits in a browser, Flow adds an agent plus character consistency and two video models, and Pomelli builds brand assets from a single website URL. All three ran without payment in the September 2026 tests, and the video model you pick inside Flow matters more than the tool you open.
Google Vids opens straight into a project from a browser, which removes the local setup that open-source video models require. Google Flow is positioned as the advanced filmmaking tool and runs Google's image and video models through an agent panel. Pomelli reverses the usual order: it reads a website first and generates branded content afterwards.
The three tools cover different jobs rather than competing versions of one product. Use the table below to pick before spending time on prompts.
| Tool | Entry point | Best for | Video model | Main limit |
|---|---|---|---|---|
| Google Vids | vids.new URL | Fast browser edits, avatars | Vids AI video | Fewest cinematic controls |
| Google Flow | Labs project | Multi-shot scenes, characters | Veo 3.1 (8s), Omni Flash (10s) | Character work needs setup |
| Pomelli | Website URL | Branded images and campaigns | Animate from a still | No prompt-first workflow |
The speaker in the source video says he tested more than ten free AI video tools, judged five useless and two weak, and kept these three. That is one person's ranking, not a published benchmark, so treat the ordering as a starting point rather than a verdict.
What Google Vids Actually Does for Free
Google Vids is the fastest of the three to start because it runs entirely in a browser and asks only for an aspect ratio before showing a grid of starting points. That grid includes recording yourself, uploading files, converting slides into video, generating AI video, and adding an AI avatar.
Orientation is the first choice: landscape, portrait or square. Landscape exposes the most controls, and the three options that carry the most weight are create AI videos, the AI avatar, and editing.
In the AI video panel you can attach reference images or paste a prompt directly. The test prompt produced a man running toward the camera with pigeons bursting upward, and the presenter describes the Notion and a short delay as the weak points rather than a failure.
Below each generated clip sit three actions: extend, edit and recreate. Mixing up the last two is the most common mistake described in the source. Recreate generates a brand-new video, while edit modifies the clip you already have. Changing a shirt to red and the scene to night kept everything else consistent, which is the point of editing a clip you already like.
The avatar path is the same idea with a presenter. Presets are previewed by hovering, a script is pasted, and play preview reads it back before you generate. Multi-scene projects let you render the current scene or every scene at once, which is the main reason to keep a longer video inside Vids instead of a single prompt.
Google Flow: Agent Prompts, Veo 3.1 and Omni Flash
Google Flow is the advanced option, and its distinguishing feature is an agent that writes and plans prompts for you. Everything generated lives inside a project, the agent is reachable from the prompt box, and the manual controls stay available if you click the model name and set parameters yourself.
Two settings absorb most of the friction. Confirm before generating, set to always, makes the agent describe the job and wait for approval before anything runs. Image defaults are set to 16:9 with Nano Banana 2 and video defaults to 16:9 with Omni Flash, while individual generations can switch to Nano Banana Pro.
The agent announces what it is about to do, and show thinking exposes its plan for the image before generation. Sending a plain-language request such as a wide cinematic image produced a still detailed enough to suggest a story, which is how the agent behaves once defaults are configured.
Video is where the model choice bites. Veo 3.1 quality runs up to 8 seconds and has no setting for cutting between camera shots, so multiple angles are written into the prompt as timed beats. Google calls this timestamp prompting: each shot gets its own time range and action, cut two separates them, and the closing lines describe what must stay constant, such as the market, the light and the characters.
The subtraction is worth noting. The presenter's clip held cinematic lighting and clean cuts, and he names realism as the weakness, saying the man does not read as a real person early in the shot.
Veo 3.1 or Omni Flash: How to Choose
Choose Veo 3.1 when shot control matters and Omni Flash when clip length matters. Veo 3.1 quality produces up to 8 seconds with no dedicated cut setting, so multi-shot scenes depend on timestamp prompting. Omni Flash reached 10 seconds in the test and kept a hand-drawn fox consistent from start to finish.
The clearest demonstration came from asking the agent for a 10-second clip of a small orange fox in a storybook style. The agent flagged the conflict itself, offering either a switch to Omni Flash or a trim to 8 seconds, which is the model-selection warning made visible.
| Model | Maximum length | Shot control | Style risk |
|---|---|---|---|
| Veo 3.1 quality | Up to 8s | Timed beats in prompt | Realism weak early |
| Omni Flash | Up to 10s | Not demonstrated in test | Consistent across clip |
The practical rule from the source is to match the model to the length you need before writing, because a long concept forced into Veo 3.1 loses detail. Neither model was measured end to end for cost or throughput, so the comparison covers clip length and prompt control only.
Character Consistency in Google Flow
Character elements are Flow's answer to characters that change between shots. In the left rail you can start from sample characters or build one from a full text description, optionally adding reference images, and the result is saved into the project rather than into a single prompt.
The build sequence is short: write the description, generate, optionally add a secondary reference shot from the body icon, give the character a name, pick a preset voice, and fill in the character info field that describes personality and behavior. Voice cloning from an uploaded recording is not available, so the choice is limited to presets.
Using the character is the part that changes workflow. Typing the at sign in the agent chat opens a picker containing saved characters, images and earlier videos, so a saved character can be dropped into a prompt like an actor receiving direction, including a spoken line.
In the test, a 10-second Omni Flash scene kept the character identical throughout and delivered the line as written. The character element solves consistency across a project, though it does not extend clip length beyond the model's own ceiling.
Pomelli: Build Branded Assets From One URL
Pomelli generates from a website instead of a prompt. You supply one URL, it analyzes the brand, and later output follows the typeface, palette, values and tone of voice it extracted. If no site exists yet, the no-website-yet path accepts up to 20 uploaded files instead.
The extracted profile is called the business DNA, and it stays editable after analysis. From there the choices are websites, a brand book, photo shoots and campaigns, and the auto-generated brand book pulls real product images from the source site rather than mock-ups.
Photo shoots start with a product and a template choice, and custom templates can be added. Six style and product templates were generated at a story aspect ratio of 9:16, and any result can be animated into video with the animate button.
Campaigns take a one-line description and return an editable brief with title, description, goal and offer before anything is produced. The presenter judged the resulting copy usable rather than generic, which is his assessment from a single run.
Limits You Will Hit With the Free Tools
Clip length, shot control and character setup are the practical limits, and all three are tied to the model rather than the tool. Veo 3.1 caps at 8 seconds and handles multiple angles only through timestamp prompting, while Omni Flash reached 10 seconds in the test without that workaround.
Vids is simplest to start and weakest on cinematic control, so it suits explainer and social clips more than scene-based filmmaking. Flow carries the agent, model selection and character elements, but a usable result depends on the one-time settings, especially confirm before generating.
Pomelli assumes a website or a file set to analyze, and it is a marketing asset generator rather than a scene-based editor. The brand DNA is editable, but the workflow begins from the brand, not from a prompt.
One framing point belongs here. The source video presents all three as free, and no pricing was verified for this article beyond that run, so check current availability before building a project around a specific model tier.
FAQ
- Which free Google AI video generators are worth using? Three tools carried the September 2026 test: Google Vids, Google Flow and Pomelli. Vids is the quickest browser start, Flow handles agent prompts, two video models and character elements, and Pomelli generates branded websites, brand books, photo shoots and campaigns from a URL.
- Is Google Veo 3.1 or Omni Flash better for free video generation? Neither wins outright. Veo 3.1 quality offers shot control through timed prompt beats but runs up to 8 seconds, while Omni Flash reached 10 seconds in the test and held a character's style across the clip. Pick the model after deciding clip length.
- How long can a Google Flow video be? In the tested setup, Veo 3.1 runs up to 8 seconds and Omni Flash produced a 10-second clip. The agent flagged the conflict when a 10-second request was aimed at Veo 3.1, offering either a model switch or a trim to 8 seconds.
- Can I keep the same character across multiple AI video shots? Character elements in Google Flow are designed for this. A character is built from a description or reference images, saved to the project, and pulled into later prompts with the at-sign picker, and the tested 10-second scene kept the character consistent throughout.
- What is timestamp prompting in Google Flow? Timestamp prompting writes cuts into a single prompt as timed beats, because Veo 3.1 has no setting for switching camera shots. Each shot receives its own time range and action, and the prompt closes by stating what stays constant across the cuts.
Skalablog CTA
The thread running through these three tools is that structure beats raw prompting: a saved character, a business DNA, a timestamped shot list. Writing has the same property, and most people already have the structure sitting in a recorded video.
If your explanation, interview or walkthrough lives in a YouTube video, you can paste the URL into Skalablog, let it transcribe the recording, and generate an article draft from the transcript. Notes, criteria and a working article tool for other creators live at crazystack.com.br, and technical walkthroughs on that channel, including work by Dev Doido do canal do youtube, are the kind of source material that converts well.
Fork this article
Start a new branch from the same video, shaped your way. You keep the credit; the original keeps the attribution.
A fork in another language is filed as a translation of this article, so the two pages point at each other. You can unlink it later from the editor.
0/240
You are creating
- Format
- For
- Language
- Source
- Your angle
You will be asked to sign in before it is generated.
Buy credits