Skip to content
How I AIHow I AI

I cloned myself with Gemini Omni in 15 minutes (and it's terrifyingly good)

In this experimental episode, I document my real-time attempt to create an AI avatar of myself using Google Flow and the new Gemini Omni video generation model. I walk through the entire process—from scanning my face with my phone to generating a complete one-minute hype video for the podcast, all in about 15 minutes. *What you’ll learn:* 1. How to create an AI avatar using Google Flow in under five minutes 2. Why video AI tools unlock creative possibilities for people with zero video production skills 3. The step-by-step process of generating a full storyboard using AI as your creative producer 4. How to use character consistency features to generate multiple video scenes with the same avatar 5. The uncanny-valley moments you’ll encounter when your AI clone doesn’t quite nail emotions or physics 6. How to stitch together AI-generated scenes into a complete video using built-in editing tools *Brought to you by:* Merge—Connective infrastructure for production AI: https://www.merge.dev/howiai Jira Product Discovery—Prioritize with insights, build with confidence: https://atlassian.com/howiai *In this episode, we cover:* (00:00) Getting started with Google Flow and Gemini Omni (01:38) The avatar creation process: scanning and photo capture (02:55) Using Flow to brainstorm a hype video storyboard (06:59) Generating the first video scene with the avatar (08:41) Troubleshooting: accidentally generating images instead of videos (09:32) Generating all seven scenes for the complete video (11:37) Reviewing the avatar videos (13:13) Stitching the videos together in the browser-based editor (14:32) The complete How I AI hype video (15:32) What worked and what didn’t (19:04) Final thoughts *Blog & detailed workflow walkthroughs from this episode:* How I Built an AI Avatar and Hype Video in 15 Minutes with Google Flow: https://www.chatprd.ai/how-i-ai/ai-avatar-video-in-15-minutes-with-google-omni-flow ↳ How to Create a Promotional Video with an AI Creative Director: https://www.chatprd.ai/how-i-ai/workflows/how-to-create-a-promotional-video-with-an-ai-creative-director ↳ How to Create a Personalized AI Avatar with Google Flow: https://www.chatprd.ai/how-i-ai/workflows/how-to-create-a-personalized-ai-avatar-with-google-flow *Tools referenced:* • Google Flow: https://labs.google/fx/tools/flow • Gemini Omni: https://gemini.google/overview/video-generation/ • Veo 3: https://deepmind.google/technologies/veo/ *Where to find Claire Vo:* ChatPRD: https://www.chatprd.ai/ Website: https://clairevo.com/ LinkedIn: https://www.linkedin.com/in/clairevo/ X: https://x.com/clairevo _Production and marketing by https://penname.co/._ _For inquiries about sponsoring the podcast, email jordan@penname.co._

Claire Vohost
Jun 3, 202620mWatch on YouTube ↗

CHAPTERS

  1. 0:00 – 0:30

    Cloning Claire with Flow + Gemini Omni: the 15-minute challenge setup

    Claire sets the goal: create an AI video avatar of herself and produce a one-minute hype video for the How I AI podcast in about 15 minutes. She frames it as an experiment that might fail, revisiting Google Flow and the Gemini Omni video model after an earlier attempt didn’t work.

    • Goal: generate a full hype video starring an AI avatar of Claire
    • Tools: Google Flow + Gemini Omni video generation
    • Acknowledges uncertainty due to past failure
    • Sets expectations for a fast, end-to-end workflow
  2. 0:30 – 1:30

    Why Merge (sponsor): production infrastructure for AI agents and integrations

    A sponsor break explains the “hidden” work of shipping AI products: integrations, permissions, routing, reliability, and cost control. Merge is positioned as the infrastructure layer to connect to many tools, enable secure agent actions, and optimize model routing/spend.

    • AI products require more than models: integrations and permissions are hard
    • Merge connects to thousands of tools and supports secure agent actions
    • Emphasis on reliability, routing, and cost efficiency in production
    • Social proof: companies like OpenAI, Dropbox, and Ramp
  3. 1:30 – 2:31

    Creating the avatar in Google Flow: QR scan, photo capture, and hoping it sticks

    Claire starts the avatar creation flow, scanning a QR code and capturing photos/head turns with her phone. She notes this previously claimed to finish but didn’t work, so she waits to see if the avatar becomes usable this time.

    • Initiates Flow’s “Create an avatar” feature
    • Phone-based capture: multiple photos + head-turn checks
    • Prior attempt failed despite “done” message
    • Wait period while Flow processes the avatar
  4. 2:31 – 4:02

    Avatar is ready: first look and immediate use-case (How I AI hype video)

    The avatar appears (including a humorous fisheye-like version), and Claire moves straight into using it for content creation. She asks Flow to help build a storyboard for a podcast hype video featuring her avatar as the consistent character.

    • Avatar successfully generated and selectable in Flow
    • Target output: hype video for the How I AI podcast
    • Uses Flow as a creative assistant, not just a generator
    • Requests multiple scenes and a clear narrative arc
  5. 4:02 – 5:33

    Creative direction & vibe: dark home office, hacker aesthetics, authentic but high-tech

    Flow prompts for creative intent (setting, tone, pacing), and Claire defines a specific look: dark green walls, AI books, posters, authentic lifestyle feel with a hacker/coding vibe. She reflects on how multimodal tools unlock video creativity she couldn’t do solo.

    • Flow gathers parameters: location, tone, pacing
    • Claire’s direction: dark home office, green walls, AI books, posters
    • Wants “authentic lifestyle” but high-tech hacker vibe
    • Broader point: AI enables non-video-creators to produce video concepts
  6. 5:33 – 6:33

    Seven-scene storyboard: keyboard close-up, office reveal, spin-chair intro, montage, CTA

    Flow proposes a roughly seven-frame structure, including a mechanical keyboard close-up, a wide office shot, a chair spin reveal, and a stylized AI montage leading to a call-to-action. Claire approves and requests the storyboard, planning to swap in her avatar via @ mentions.

    • Storyboard includes ~7 scenes with escalating energy
    • Key beats: typing, environment reveal, character reveal, montage, mic/CTA
    • Claire requests a generated storyboard and plans to reference @me avatar
    • Notes early issues with Flow reliably referencing the avatar
  7. 6:33 – 8:34

    Storyboard images arrive—without the avatar—so Claire manually adapts prompts

    The storyboard grid generates, but Flow can’t apply the avatar within the storyboard itself. Claire likes the visuals and decides to copy a frame prompt and replace “Claire” with the @me avatar reference to generate video scenes featuring her character.

    • Flow generates storyboard visuals but not with the avatar
    • Claire evaluates the style and selects scenes to produce
    • Manual workflow: copy prompt + replace name with avatar handle
    • Goal: consistent character videos using the avatar reference
  8. 8:34 – 9:34

    First generation attempt: accidental image output, then switching to video mode

    Claire initially generates still images by mistake due to the image/video toggle. She corrects the setting to video generation and re-runs the prompt, noting that video takes longer and typically returns two variants per scene.

    • Common pitfall: wrong mode selected (image vs video)
    • Re-prompts with avatar and scene description (hands typing, lighting, camera)
    • Video generation produces two versions (V1/V2)
    • Plans to queue additional scenes while waiting
  9. 9:34 – 10:35

    Early results: background leakage from training photos + spooky realism

    Claire notices the model reuses elements from her real environment—posters/books behind her during capture—inside the generated video. The first video outputs are impressive but eerie, reinforcing the “clone” feeling and motivating her to generate all scenes for a full montage.

    • Avatar capture context influences generated backgrounds
    • Eerie/spooky effect of seeing ‘AI Claire’ move and speak
    • First outputs show unexpected details (e.g., nails, props)
    • Decision: generate all scenes and assemble into a complete hype video
  10. 10:35 – 11:36

    Why Jira Product Discovery (sponsor): aligning teams on what to build

    A second sponsor segment focuses on the challenge of ‘multiplayer’ product decisions: scattered docs, stale roadmaps, and misalignment. Jira Product Discovery is framed as a shared system for capturing ideas, prioritizing, and handing off execution into Jira.

    • Problem: alignment and decision-making across teams
    • Centralizes ideas, prioritization, and living roadmaps
    • Uses Atlassian teamwork graph to suggest what to build next
    • Smooth handoff from decision to delivery via Jira
  11. 11:36 – 13:39

    Generating all seven scenes: jump-scares, chair spins, and choosing the best takes

    With seven scenes running, Claire reviews outputs—especially a spinning-chair ‘intro’ where the avatar pushes glasses she doesn’t wear. She compares the two versions, noting background oddities and selecting the take that looks best for the final edit.

    • Seven scenes generated to match the storyboard order
    • Notable scene: chair spin reveal; comedic/uncanny variations
    • Two versions per scene enable quick selection of best take
    • Notices surprising background artifacts (e.g., ‘NVIDIA Way’)
  12. 13:39 – 14:41

    In-browser editing in Flow: stitching scenes into a single timeline

    Claire opens Flow’s built-in browser editor and assembles the chosen clips in storyboard order. She emphasizes how little time it took—capturing the avatar, generating scenes, and editing—before previewing the final hype video end-to-end.

    • Flow includes a timeline editor directly in the browser
    • Simple assembly: drop in chosen clips in recommended order
    • End-to-end process reportedly fits under ~15 minutes
    • Prepares for the full reveal of the finished video
  13. 14:41 – 15:16

    Final reveal: the completed How I AI hype video

    Claire plays the finished montage: a dramatic opening (“We were told AI would replace us”), her on-camera intro as the avatar, and a call-to-action to subscribe. The result is compelling, slightly glitchy, and surprisingly coherent for such a fast workflow.

    • Full playback of the stitched hype video
    • Includes narration, identity reveal, and subscribe CTA
    • Noticeable quirks: timing overlaps and minor glitches
    • Overall: shockingly strong output for rapid generation
  14. 15:16 – 18:48

    Postmortem: what worked, what didn’t, and where Omni/Flow still fall short

    Claire breaks down strengths (speed, accessibility, surprisingly accurate likeness at times) and weaknesses (uncanny valley expressions, inconsistent hair/background, cheesy ‘AI’ visual tropes, weak typography/graphics). She concludes that with better prompting and more reference images, it could become highly convincing.

    • Pros: fast creation, low expertise required, sometimes very face-accurate
    • Consistency issues: hair length, background elements, lighting drift
    • Uncanny emotions and awkward moments (laughing looks off)
    • Graphics/typography not strong; AI aesthetics feel dated/cheesy
  15. 18:48 – 20:35

    Closing thoughts: obsession, next experiments, and viewer call-to-action

    Claire says she’s blown away and wants to keep experimenting with Flow and the Omni model as a new hobby project. She invites viewers to try creating their own avatars, share results in comments, and supports the show with standard like/subscribe and podcast platform plugs.

    • Plans to spend more time refining prompts and consistency
    • Invites audience to test avatar creation and share examples
    • Reflects on the episode as a ‘How I AI’ success story
    • Standard closing: like/subscribe, ratings, and where to find the show

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.