
@kaiganii
iAccount based inUnited States
About this account
- Account based in
- United States
- Connected via
- United States App Store
Account-level information from X, not a live location or the device used for a specific post.
Gen AI Video Creative Director & Founder, prev at Sony Pictures and Netflix. Co-host of 5ives podcast. CPP: Hailuo Minimax, OpenArt, Luma, Dreamina, Vidu
Los Angeles, CA
Joined April 2007
- Tweets42.1K
- Following5.7K
- Followers14.2K
- Likes9.2K
Pinned Tweet
FULL VIDEO: I edited together close to 120 ChatGPT images into video clips made in
@Hailuo_AI
image-to-video to match the Dune trailer frame-by-frame in the style of Moebius
Our latest 5ives retrospective on Gossip Goblin.
youtube.com/watch?v=S95LLFdsβ¦
@VanFalk_
I really hate how this current generation of AI models will bury what your looking for inside a path like...
I've started rebuilding the CineRecon benchmark in ThreeJS as CineReconJS.
kaigani.github.io/CineReconJβ¦
h/t to @fonkydonk / @AIForHumansShow for inspiring us to try a new documentary format.
Who is Gossip Goblin β and can his strange AI worlds make the leap to the big screen?
Our first 5ives Retrospective follows Zack London through his filmmaking process and Gods Don't Give Gifts.
WATCH NOW: youtu.be/S95LLFdsagQ
@VanFalk_ @Gossip_Goblin @fableshowrunner
Who is Gossip Goblin β and can his strange AI worlds make the leap to the big screen?
Our first 5ives Retrospective follows Zack London through his filmmaking process and Gods Don't Give Gifts.
WATCH NOW: youtu.be/S95LLFdsagQ
@VanFalk_ @Gossip_Goblin @fableshowrunner
Updated by AI avatar ... turns out I've been using a backrooms version of myself all this time!
π€ Made with AI
I didnβt expect OpenAIβs Dots to be good with kids but it was joking with her and making her laugh.
Cool project benchmarking how well AI models reconstruct video scenes in Blender. (I might have to build a version of this using ThreeJS)
cinerecon.github.io/
I created a review panel of the top frontier models and then had them score each other's performance.
I've been developing a system that explores the styles that AI video and image tools are able to produce and then gives you a short list in response to a brief.
My favorite old PC game (used to play this to relax until I switched back to Mac)
β
Decompiled & rebuilt from .exe to run on MacOS natively
β
Graphics upscaled in gpt-image-2.5
β
Midi tracks remastered as mp3 in YuE2
Edging closer & closer to Black Mirror
Chinese streamer Fanqie Jiang Ye found an interesting way to get donations from viewers, charging $7.50 to make her spin π
- many viewers repeated the donation multiple times in a row causing her to spin nonstop.
I've collected a lot of examples of animation styles for my research β so I thought, why not point AI at it and make a documentary.
youtu.be/gOKbF43zk0Y
I think Taruma is an underrated genius in the GenAI space, and you should study his "auteur script" closely. It's made a huge difference in the quality of my results.
SceneFlow Update: Introducing the Multi-Track Timeline & Studio Workspace.
SceneFlow isnβt an AI web app that generates video or scripts. Itβs an interactive player and alignment workstation where you bring your video and prompt together to verify what the model followed, what it missed, and adjust timing.
Over the past week, I rebuilt the two areas that caused the most friction: watching dense scenes without card clutter, and editing cue timings without fighting the player.
Here is whatβs new:
π Multi-Track Sync Timeline β Active cues now glide along dedicated horizontal tracks (Dialogue, Camera, Audio, Action, VFX, etc.). Overlapping events no longer fight for screen space or clutter the view.
π Filters & Zoom Presets β Mute or isolate specific categories in one click, and switch between 4s, 8s, and 16s zoom horizons to inspect rapid dialogue or step back for overall scene pacing.
π οΈ 3-Panel Edit Workstation β Rebuilt Edit Mode into a 3-column studio layout. Switching between Playback and Edit no longer resets your video to 0:00 or cuts audio.
π Live Loop Preview & Micro-Nudge Steppers β Adjust cue start/end times with Β±0.1s and Β±0.5s steppers while a continuous loop plays in real time. Tweak boundaries by ear until the cut lands right on the beat.
π€ Built-In Gemini Prompt & Schema β Need AI to help draft timestamps? Copy the official multimodal prompt and schema directly from the cues window, run it in Google AI Studio, and paste the JSON straight in.
π Studio Script Editor β A revamped 3-panel screenplay editor with an automatic section outline, word wrap that keeps line numbers strictly aligned, and a 1-click syntax guide.
π¬ Built for the "Auteur Script" β The layout and syntax are designed around my personal prompting framework (structuring scenes with intent, logic guardrails, and chained -> execution states). Hopefully, my next article will dive deep into this!
π Read the New Deep-Dive Article: taruma.substack.com/p/scenefβ¦
Links below π




















