AI Filmmaking

Create cinematic AI videos with Google Flow and Gemini Omni

Create your own avatar, plan a consistent short film, refine clips with chat, add cinematic audio and try video-to-video editing—a practical companion to the Hindi masterclass.

By Dayanand Vishwakarma · 9 min read

Watch this tutorial on YouTube ↗

Turn the workflow from Dayanand Vishwakarma’s video into a project you can follow one step at a time. You will start with your own avatar, test one short clip, then build a small film from approved references and scenes. The copyable prompts below preserve or adapt the examples in the supplied script; extra instructions are labeled. Results, feature access and credit costs vary, so begin with a small test.

WHAT YOU WILL MAKE

One avatar test, a short sequence of consistent story clips, a cinematic edit and an optional fantasy effect applied to your own recording.

Before you start

Access to Google Flow and the relevant features in your account and region.
A phone with camera and microphone for your own avatar setup.
A computer, a supported browser and a stable connection.
A short original or permission-cleared script and references.
Available generation credits; review current costs before each batch.
Premiere Pro or another editor for assembling the final film.

01
Open Flow and plan a small first project
Open Google Flow from the official link below and create a new project. Use a computer for the editing workflow, with your phone nearby for avatar recording. Name the project and create folders for References, Script, Generated Clips and Final Edit.
Start with one short test rather than an entire film. Check which models, avatar features and generation options your account currently offers. Access can vary by region and plan. The video demonstrates Omni Flash; select the compatible Omni model shown in your account.
Important cost note: this workflow uses generation credits. The video mentions a Jio/Gemini promotion and specific credit prices from its demonstration; these are not a permanent free-access promise. Check your current account allowance and any offer terms before proceeding.
02
Record your own avatar
In the video, Dayanand opens the avatar option from the project interface. If your layout differs, look for Create avatar in your profile menu. Select Get started and use the displayed QR code or phone link to continue on your own mobile device.
Use your own face and voice. Read the consent screen yourself, then follow the requested camera and microphone steps. Record in good, even lighting with little background noise. Say the displayed words or numbers clearly and turn your head only as the on-screen instructions request.
Return to the computer when setup finishes. Check the avatar preview before generating. Keep the recording private and use the avatar only in contexts you are comfortable publishing. If the option is missing, consult the official avatar availability guide below rather than uploading someone else’s likeness.
03
Make a cloud-vlogging test clip
Add your avatar to the prompt using the avatar selector. The video shows typing @ and choosing the avatar; the current interface may offer @me. Confirm the correct avatar is attached, rather than only writing a person’s name.
Copy the cloud-vlogging prompt below. It separates English action instructions from the Hinglish dialogue used in the video. Choose 16:9 to follow the demonstration, an available short duration, and the compatible Omni model. Start with one output to review before generating variations.
Check the displayed credit cost and approve only the generation you want. Play the result and compare the face, voice, mouth movement and hands with your reference. Identity and voice consistency can vary; a good demo result is not a guarantee that every clip will match perfectly.
TRY THIS STEP
Vlog above the clouds
Action and dialogue from the supplied video script. Attach your own avatar before generating.
Copy the prompt, replace any [PLACEHOLDERS], and use it in your AI tool.
04
Write a story before generating scenes
Choose one short story with a clear beginning, problem and resolution. The video uses a story inspired by Premchand’s Poos Ki Raat. You can instead write an original story or use material you have permission to adapt. Do not assume a modern translation or adaptation has the same rights as the original.
Use a writing assistant of your choice, or write the script yourself. The story prompt below is an expanded English adaptation of the instruction in the video. Read the dialogue aloud and cut anything that cannot be spoken naturally within a short shot.
For your first attempt, a 30–60 second demo is easier to evaluate than a three-minute film. The video plans a 180-second film but shows an approximately one-minute edited demonstration. Keep the planned duration and the actual final edit distinct.
TRY THIS STEP
Write your short-film script
Expanded English adaptation of the story-writing instruction. Replace the story and duration first.
Copy the prompt, replace any [PLACEHOLDERS], and use it in your AI tool.
05
Turn the script into a shot-by-shot plan
Break the approved script into short clips supported by your selected model. The video works in roughly 8–10 second segments. A 180-second draft with ten-second segments would need about 18 clips before trimming, pauses or transitions; it is a planning estimate, not an automatic final runtime.
For each shot, write the scene number, duration, location, action, camera framing, English visual instructions, Hindi dialogue and sound. Keep a separate character reference sheet listing face, hair, clothes, body proportions and distinguishing details.
Use the scene-planning prompt below. Check that every scene can work as one short clip and that character and location details remain consistent. A long paragraph with many unrelated actions is harder to control than one clear shot.
TRY THIS STEP
Break the story into short shots
Expanded adaptation of the scene-splitting instruction in the video. Paste your approved script after this prompt.
Copy the prompt, replace any [PLACEHOLDERS], and use it in your AI tool.
06
Configure the Agent and approve references
If you want Flow to help organize generation, enable Agent and open its settings. Keep Confirm before generating set to Always so you can review proposed credit spending. Choose 16:9 to match the tutorial, a suitable image model and an available compatible video model.
The video uses Nano Banana Pro for images and Omni Flash for video. Use those options if available, or check the current supported-features guide before selecting alternatives. Review output count and duration as well as the model name.
Paste the Agent brief below, followed by your approved script and character descriptions. Ask it to prepare characters, locations and a shot plan first. Approve those references only when they match your story; otherwise request a specific change before spending credits on clips.
TRY THIS STEP
Brief the Flow Agent
English adaptation of the Agent instructions, with an added plan-and-approval step.
Copy the prompt, replace any [PLACEHOLDERS], and use it in your AI tool.
07
Generate in small batches and check continuity
Ask the Agent to generate the first small batch. Read its proposed shots and the credit total before approving. The video shows a four-clip batch quoted at 60 credits; your current model, resolution and settings may produce a different price.
Review every clip for face, body, outfit, location, dialogue and camera continuity. Compare adjacent shots side by side. When one fails, name the exact problem and request a revised version of that shot rather than regenerating the whole film.
If a scene is blocked, read the reason and revise the story to comply with the tool’s rules. Keep the intended safe storytelling purpose; do not try to evade its protections. Download approved clips with ordered filenames such as scene-01.mp4 and scene-02.mp4.
08
Refine one clip through chat editing
Select the generated clip you want to change. Use the edit box, or drag the clip into the Agent prompt box, and describe one clear change. The example below adapts the rain-and-neon edit from the video.
Keep the original version so you can compare the edit. Request preservation of the subject and framing, then check that the model actually did so. Follow-up edits can alter details you did not intend to change.
Look particularly at the face, wet surfaces and rain direction. In the demonstration, Dayanand notes that rain can look unnatural on the face. If that happens, reduce the intensity, simplify the request or keep the better original take.
TRY THIS STEP
Add rain and neon atmosphere
Adapted from the multi-turn editing example, with an added preservation instruction.
Copy the prompt, replace any [PLACEHOLDERS], and use it in your AI tool.
09
Add a cinematic grade and native audio
Attach your selected clip and use the cinematic-look prompt below. It follows the video’s dark teal-and-orange grade, suspenseful music and rain sound request, with additional instructions to keep dialogue understandable.
Preview visuals and audio together. Listen for clear dialogue, abrupt sound changes and background music that overpowers speech. Check that the atmosphere fits the story rather than adding dramatic effects to every scene.
Native audio and edit capabilities depend on the active model and account features. If the tool cannot produce a usable soundtrack, finish the sound in your editor using audio you are allowed to use. Keep required AI-origin labels and watermarks.
TRY THIS STEP
Give the clip a cinematic look and sound
The grading and audio request from the script, expanded to protect dialogue clarity.
Copy the prompt, replace any [PLACEHOLDERS], and use it in your AI tool.
10
Try video-to-video editing with your own recording
Record a short, steady clip of yourself holding a phone. Keep your face and phone screen visible, use even lighting and avoid fast camera movement. Upload only a clip you have the right to edit.
Drag the recording into the supported video-editing workflow. Use the fantasy-effect prompt below, adapted from the snake-emerging-from-a-phone example in the video. This is a fictional visual effect; no real animal or dangerous physical setup is needed.
Review the result frame by frame: the face should stay recognizable, the phone should keep its shape, and the effect should appear attached to the screen. Use a shorter, simpler source clip if the model struggles. Feature and input-length limits can change; check the current model settings.
TRY THIS STEP
Create a fantasy phone effect
Expanded adaptation of the video-to-video example. Use your own recording; the effect is fictional.
Copy the prompt, replace any [PLACEHOLDERS], and use it in your AI tool.
11
Assemble the film and export
Download your approved clips and import them into Premiere Pro or an editor you know. Place them in script order, trim unwanted starts and endings, and match the timing of the dialogue between shots.
Use simple cuts first. Balance audio levels, add captions if useful and verify their spelling. Match the project aspect ratio to the intended release: use 16:9 for the tutorial’s horizontal film, or plan and review a separate vertical edit.
Export an MP4 supported by your destination platform and watch the exported file, not just the editor preview. Keep the script, prompts, references and approved clips together so you can revisit the project.
12
Review, label and improve the next version
Before publishing, check facial consistency, hands, clothing, locations, rain artifacts, lip sync, audio clarity and the continuity between scenes. Fix the weakest shot first. A short coherent demo is more useful than a longer film with obvious mismatches.
Make the AI-generated or AI-edited nature clear where appropriate and follow your platform’s disclosure requirements. Use only material and likenesses you are authorized to use.
Watch the original video again to compare the demonstrated interface with your account. Model names, promotions, credit prices and features evolve. Save your working settings and the actual credit use so your next project is easier to plan.