← The AI Hype Audit — all 100 verdicts
PARTLY "5 open-source AI video generators — the same tech behind Runway and Kling, free on your machine" — a claim that names zero of the five
The claim5 open-source AI video generators most people are sleeping on — the same kind of tech behind paid tools like Runway, Kling, and Higgsfield, except free and running on your own machine.
The category is real and 2026 was genuinely its year: Wan 2.2 (Apache-licensed, Alibaba), HunyuanVideo 1.5 (Tencent), and LTX-Video produce output that independent comparisons put near mid-tier paid platforms. But this caption names none of its five repos — the list lives only in the slides, so the claim can't be pinned down as posted, and that alone caps the grade. 'Same kind of tech' is defensible (same architecture family); 'the same tech' would be false — Runway's and Kling's actual models are proprietary. And 'on your own machine' silently means a 16–24GB-VRAM GPU: a $1,600+ graphics card and minutes-per-clip render times. The paid tools' real product is speed and not owning the hardware.
What holds up
- The open-source video-gen class is real: Wan 2.2, HunyuanVideo 1.5, LTX-Video, Mochi, CogVideoX are the standard roster and several carry permissive licenses
- Independent comparisons put the best open models near Kling-1.5-level on some dimensions — the gap genuinely narrowed
- Higgsfield (name-checked as a paid tool) is itself largely a product layer on other models — see our entry #062
What doesn't
- Zero repos named in the caption — unfalsifiable as posted, and unusual for this creator's format
- "Your own machine" assumes a 16–24GB GPU most viewers don't own
- Top paid tools still win on prompt adherence and polish — 'sleeping on' implies parity that isn't there yet
- Caption funnels to the creator's paid 1:1 calls and app builds
The catch
You're saving a claim, not a list. When the names are only in the slides, the post is optimized for the save button, not for your ability to check it.
How to actually do it
- The actual list worth knowing: Wan 2.2, HunyuanVideo 1.5, LTX-Video — all with public repos you can inspect
- Own a 16GB+ GPU? ComfyUI + Wan 2.2 is the best-documented local path
- No GPU? Rent one: RunPod/Vast at roughly $0.40–0.80/hr for a 24–48GB card — cents per clip, no subscription
- Expect minutes per clip locally vs seconds on paid platforms — that speed IS what the subscription buys
- Keep the paid tool if video is client work; use open models for volume and experiments
The named version of the claim, plus the no-GPU route.
- Effort
- A day to a working ComfyUI pipeline if you're technical
- Cost
- $0 models + a $1,600+ GPU you may not own, or ~$0.50/hr rented
- Stack
- ComfyUI + Wan 2.2 (local) or a cloud GPU by the hour
- Confidence
- Medium
- Posted by
- an AI-content creator (a real engineer) who sells 1:1 consults and app builds
We test hype for free. We build the real thing for a living.
Thirty minutes, no pitch — and you'll leave with something useful either way.
Book a call with Todd or start with the free Business Checkup →The Verdict Weekly
Three verdicts every Friday. Free forever, unsubscribe anytime, no spam — that would be ironic.
© Schreier Group · schreiergroup.com · See a wild AI claim? Drop it here and we'll test it.