JimVids is a local web app that researches, scripts, voices, animates, and assembles faceless YouTube videos — plus 6 vertical Shorts every time. Powered by your existing ComfyUI + LTX 2.5, local Qwen, and ElevenLabs. No cloud bills. No subscriptions.
No cherry-picked mocks — this is an actual test render from the pipeline's scene-review demo: intro title card, ducked background music, word-timed captions, stitched from approved scenes only.
Qwen picks the 6 best standalone hooks — frame-blurred backdrop, hook captions, ready to post.
Every stage snapshots its state — crash, reboot, or OOM? Resume with one click right where you left off.
Scripts are constrained to ~8–11s scenes. Any narration over 12s auto-splits into consecutive visuals. No boring talking slides.
Qwen LLM calls and video renders share a mutex so they never collide. ComfyUI models are unloaded between every render. Real 24GB discipline.
Don't like clip #7? Reroll only clip #7 with optional prompt tweaks. Approved scenes stay untouched, downstream stitch re-runs cleanly.
Pick your timeframe before generating. Videos 8m+ switch to chapter mode: AI writes an outline first, then each chapter with continuity context.
Default: pipeline pauses after clips render so you approve scenes. Or tick auto-approve and wake up to a finished video.
Not random crops — Qwen picks the 6 best standalone hooks with captions and blurred-backdrop framing. Double the reach per topic.
Windows or Linux. No Docker. No cloud account except ElevenLabs.
| Component | Requirement | Notes |
|---|---|---|
| GPU | RTX 3090 (24GB) or similar | Ampere+ recommended |
| ComfyUI | 0.2+ with LTX 2.5 | uses your existing setup |
| Local LLM | Qwen 2.5 (Ollama/LM Studio/vLLM) | OpenAI-compatible server |
| TTS | ElevenLabs API key | free tier works |
| FFmpeg | 6.x on PATH | winget / apt |
Zero. LLM inference and video rendering run on your own GPU. The only external API is ElevenLabs for voice — its free tier gets you through several videos per month, then a few dollars a month if you scale.
Yes. Use a free Cloudflare Tunnel so the dashboard appears at a private subdomain like app.jimvids.com behind sign-in. Your RTX 3090 at home does the renders while you queue jobs from anywhere.
JimVids has a built-in mock/simulation mode (python run.py --mock) that produces real MP4/MP3/SRT assets with zero GPU — test the entire UI flow in 60 seconds.
A packaged export folder + zip: the stitched long video, 6 vertical Shorts, word-timed SRTs, intro title card, clickable YouTube thumbnail, title/description copy, and manifest JSON.
Download once, render forever. Built for creators who prefer to own their tools.
no spam • unsubscribe anytime