AI video creator and editor, Israel

Short AI videos, end to end: script, generation, voice, captions, cut and upload. Hebrew and English. Based in Tel Aviv, working remote or on site.

60 second reel. Sound on.

Work

Each card is one job: the problem, what I made, the tools, the result. Numbers come from the files, not from memory.

Problem: An AI question with a real answer takes minutes to read and nobody reads it.
What I made: 25 to 60 second Shorts: one question, many AI models, the numbers on screen, a voice that keeps pace.
Tools: OpenRouter evals, Python, ElevenLabs, Whisper checks, ffmpeg, Playwright frames
Result: 15 Shorts on one channel in two weeks, 5.6K views, best Short 1.3K. Script to posted Short in under 1.5 hours.

Problem: Hebrew AI content is thin, and Hebrew captions break in most tools.
What I made: A Hebrew Short on who the AIs would vote for, with burned right-to-left captions and a Hebrew voice checked word by word.
Tools: Claude Code, ElevenLabs Hebrew voice, niqqud respelling, Whisper large-v3, ffmpeg
Result: 86 models, 4,191 answers in one eval. Eight cuts in one day to lock the final.

Problem: A character has to look, sound and move the same in every episode, or people stop believing it.
What I made: A satire influencer with a triangle head: script, voice, stills, talking clips, lip-sync, Hebrew captions, 13 to 30 second episodes.
Tools: Gemini TTS, nano-banana, Kling 4.0, sync v3 lip-sync, ffmpeg
Result: Six locked episodes, first one live on Instagram, YouTube and Facebook. About three dollars of API spend per episode plus Kling credits.

Problem: The club wanted a film look for Reels and ads without a shoot day.
What I made: A 38 second vertical film: stills in the club's look, motion from the stills, a voice, music, type in the brand fonts, cut on the beat.
Tools: Higgsfield Soul 2, Kling v3 and 4.0, ElevenLabs voice and music, Scribe timing, ffmpeg at 118 BPM
Result: Three client rounds. The AI footage was approved in writing. Delivered with and without captions.

Problem: A sponsor wanted a presenter who posts daily, without a studio or an actor.
What I made: One face and one voice from a single still image. Market cards and story clips, vertical and 16:9, with a check that the spoken number equals the number on screen.
Tools: Higgsfield Soul, nano-banana, Kling AI Avatar, Chatterbox voice, Python composer with QC gates
Result: 21 draft videos delivered. Under $2.10 per video. Release waits on sponsor consent.

Problem: Posting by hand eats the day and mistakes go live.
What I made: One Python system: voice, render, loudness and caption checks, a numbers check, per-platform caption packs, YouTube upload with a channel check, a scheduler and a watchdog.
Tools: Python, ffmpeg, YouTube Data API, Metricool, Upload-Post, launchd
Result: 1,092 automated tests. Nothing posts without a one-click approval.

Problem: An international campaign needed explainer videos and screen animations fast, on a fixed budget.
What I made: Two 16:9 AI map videos and about 25 animated signage files in odd screen sizes.
Tools: Seedance, Kling, ElevenLabs, After Effects, ffmpeg, a budget-gated job runner
Result: Delivered and signed off. Earlier work for the same market: commercial content broadcast on Israeli TV.

Problem: Cloud generation adds up when you iterate a hundred times.
What I made: A 30 second reel with one invented woman kept consistent across eight scenes, made entirely on a Mac.
Tools: Draw Things, FLUX.2 klein, LTX-2.3, local Whisper
Result: Zero dollars in services. The lessons went into every paid job after it.

Two ways to start. Quotes come after a short call, per project or per month.

Start here

Ongoing

Say what you need and which platform it goes to. I write back myself.

Or write directly: mail@yuvaldor.com