I2V Studio – An automated AI video tool powered by Google Flow

The creation screen follows five clear stages: pick a chatbot (script template), pick a topic, review the script, build the video, publish. The first two stages cost no Google credits, so you can explore freely before committing.

At the topic stage the app opens that template's chatbot with your Google account and asks what is worth covering today — usually answering within about half a minute. You pick one of the topics it returns or type your own, force a length between 30 and 180 seconds, and declare a voice for each character so every scene speaks in the same voice.

The script comes back split into scenes, each with its own narration, an English prompt for the video model and its own timecode. This is the checkpoint before credits are spent: you read every scene, edit the narration, edit the prompt, add or remove scenes, or ask the AI to rewrite the whole script.

Clip generation runs on Google Flow with four video models: Omni 1.1 Flash (4, 6, 8 or 10 second clips, 7–15 credits per run), Veo 3.1 Lite (10 credits), Veo 3.1 Fast and Veo 3.1 Quality. Choosing the model is up to you, and the app shows the price before you commit.

Not every template follows exactly those five steps. Product templates replace the topic suggestions with a form where you describe your product; templates with several treatments insert an extra step for choosing one. The progress bar changes per template instead of forcing every project into the same shape.

For image-based templates, step three is called “Review script & images” and splits in two. The first stage generates a still image for each scene with Nano Banana on Google Flow — the app states plainly that this stage does not deduct credits, so pick the best image tier and regenerate until you are happy.

You approve the image set in front of you before moving to the second stage: choose a video model, see what the app says it will cost in credits, and only then turn the images into clips. What you pay credits for is a frame you have already seen and agreed to, not a gamble.

Editing a scene's image prompt discards both the old image and the clip built from it — the app says so right under the input instead of letting you discover it through a credit bill.

The panel on the right sums up the batch before it runs: how many scenes, how many images are about to be generated, and what it costs. Scene images are also generated in a chain so the setting and lighting do not jump from one scene to the next.

Every project opens a workbench: a preview frame for the selected scene, a scene timeline, and a right-hand column for editing that scene's content, clip, audio and subtitles.

This is where most credits are saved. When the model gets one scene wrong, you edit that scene's prompt and regenerate it — the app shows what the regeneration will cost. The other scenes that came out well stay untouched; nothing is rebuilt from scratch.

Narration, voice and subtitles are editable in the same place. Change the voice for one scene, fix a subtitle line the model misread, then export again — the parts you did not change are not rebuilt.

The timeline shows which scenes already have a clip, which do not, and which have subtitles. You can duplicate a scene, add a new one or delete a spare one before publishing.

The template library sits right in step 1 of the creation screen. Each template is its own scriptwriting brain — a persona, a script blueprint and its own artwork — one per video style: product reviews, animated characters, stick-figure storytelling, POV life hacks, book recommendations, virtual fitness coaches and more.

The grid is split across nine industries, each chip showing how many templates it holds, and the search box accepts Vietnamese typed without diacritics. Tap a card to preview the kind of video that template produces — tap again to stop. Only once you have picked one does the app go and ask the chatbot, and none of this costs a single credit.

The Customise template button opens a panel that spells out how that template builds a video before you commit: how many seconds per scene, whether the video keeps the audio Veo generates or gets a voice-over, whether each scene opens on the previous scene's last frame, the image-style block appended to every prompt, and the example library the chatbot learns from. Seven templates can pin a character with a reference photo so the character keeps the same appearance across videos.

The contents of a built-in template ship with the installer and deliberately cannot be edited: change a single example prompt and the whole template's script blueprint drifts, and the next update would overwrite it anyway. In exchange, updates usually add new templates to the library — install one and they are simply there, at no extra cost. What belongs to you is each template's character photo — it lives in the data folder on your machine, so installing an update never touches it.

Each template has its own customisation panel, and the most valuable thing in it is the channel's character image. The app attaches that image to every scene while generating clips, so all scenes show the same person. Without it, Google Flow reinvents a face from the text for every scene — the clothes match, the face drifts.

The character image is generated inside the app and costs no Google credits: the app opens Flow with your account and asks for a full-body portrait built from the character description itself. For a closer match, upload your own image — standing straight, facing the camera, plain background, nobody else in frame.

The character description is written in English and copied verbatim to the front of every image prompt. Editing it also rewrites the prefix inside the template's example prompts, because the chatbot learns from examples rather than instructions — leave the old examples in place and your edit simply does not take.

The same panel holds the image style — a line appended to every image prompt so all scenes share one look — along with the template's scene pacing and how audio is handled. This is where a built-in template becomes your own channel.

Edge is the default source: 8 voices across Vietnamese, English, Japanese, Korean and Chinese, free, nothing to sign up for, and it reports word-level timings so subtitles line up exactly.

CapCut adds 24 Vietnamese voices with character — news-reader, film-review, storytelling — instead of flat narration. It is also free with nothing to sign up for, but it goes through an unofficial route, so the app says plainly that it may stop working.

Piper and VieNeu run entirely on your machine: press Download once per voice (20–63 MB) and afterwards you need no network and nobody counts your usage. The trade-off is that neither returns timings, so subtitles are spread evenly by word length — slightly off in sentences that mix short and long words, and the interface says so instead of hiding it.

ElevenLabs gives the best voice quality, uses your own API key and bills you per character by ElevenLabs. The key is stored in the data folder on your machine and is never shown back in full. A voice that is not ready is blocked at confirmation time rather than at assembly time — so a whole batch of render credits is not lost to a video that dies at the last step.

You add your own Google accounts. A blank Chrome window opens, you sign in as usual, and the app notices the moment you are done and closes the window for you. It never touches your password — it only opens the window and waits.

Each account gets its own Chrome folder, so several accounts can run in parallel: measured on Windows 11, 8 Chrome windows took 8.2 seconds to start, each takes 0.6–1 GB of RAM, and a 16 GB machine handles roughly 13 at once.

Projects, the licence and exported videos live in the app's data folder on your machine, not next to the executable — so installing over the top during an update leaves your work alone. The export folder can be changed in Settings.

The licence is activated against a machine code: buy once, use forever, and when you change machines you release the licence on the old one and activate on the new one from inside the app. Updates download straight from the licence server with a checksum check.