A discussion thread on Product Hunt around Gemini Omni Flash is starting to attract attention in AI underground circles, and the signals are worth noting BEFORE this lands in the mainstream tech press.

Google has positioned Omni Flash as a fast, multimodal model capable of generating and editing videos through a conversational interface. That sounds promising — and some use cases actually work reasonably well. But there's a fairly long list of practical problems that many users only encounter after they've already paid.

The most concrete issue is price versus reality. The model costs approximately $0.10 per second of output, which ostensibly yields a 10-second clip for $1. That's technically accurate. But because quality varies and you typically need 4–6 attempts per usable clip, the real cost is closer to $6 per clip you can actually use for something. On top of that, users on cheaper plans ($20/month) report hitting daily caps after just 3–5 videos.

You're paying for experimentation, not for output — and that's not what people thought they were buying.

Beyond pricing, there are technical shortcomings generating noise as well. Text rendering in videos is poor — signs, subtitles, and labels degrade significantly. After three or four rounds of editing within a conversational session, characters and scenes begin to lose consistency. And audio is limited to voice-over narration; no background music, no sound effects.

Gemini Omni Flash: Google's video model has hidden costs users are furious about - Bilde 1

Worth noting: research signals indicate that Gemini 3 Flash (a close relative) showed a 91% hallucination rate in one benchmark. That's an alarming figure, and while Omni Flash is a different variant, it's reasonable to be skeptical about the factual accuracy of output from the Flash family in general.

What does this mean? First, this is still public preview, and Google is open about the fact that longer clips and improvements are on the way. They've already doubled the video limits for Ultra subscribers after a bug consumed too many credits. So there are signs they're listening.

Second, this is an early warning signal for anyone considering Omni Flash for creative workflows in 2026: build in a solid buffer for both budget and timeline. The model is interesting as a technological direction — conversational video editing is genuinely exciting — but it is not production-ready for anything requiring consistency, on-screen text, or more than a handful of clips per day.

Stay tuned. These are early community signals, not a final verdict.