Gemini felt dumb on December 23, 2025.

I was skeptical about switching from ChatGPT, but after trying Gemini 3 Pro and Flash 3 I was blown away. The model instantly understood my prompts, answered with depth, and was lightning‑fast. It felt like a new layer of intelligence that made my daily chats smoother and more productive—so impressive I can’t imagine going back.

I tried using Gemini to draft press releases and kept feeding it YouTube links as source material. Every time it says it’s “looking” at the video, then just hangs and finally spits out a generic error asking to try something else. No matter which video I pick, the YouTube‑to‑text pipeline just fails, while the same prompt without a link works fine. This constant breakage is really frustrating.

I kept generating solid images in Gemini, but every time I tried to download one, a huge “Error occurred while attempting to download the image” popped up. It’s random, sometimes the images stay low‑quality and then vanish entirely. I’m stuck watching my work disappear, and the whole thing feels unbelievably frustrating.

I tried Google Veo 3.1 through Flow to see if it could out‑perform Kling for the “celebrity selfie” trend. By generating a start‑frame selfie and an end‑frame with a laugh, then letting Veo interpolate the motion, I noticed the lighting stayed impressively steady on the face while the head turned. Kling’s results were wilder but the light shifted oddly. The tutorial I posted shows the prompts and settings, and I’m curious what others think of Veo’s consistency.

I was really excited to try Gemini 3 Pro/Flash, but it turned out to be one of the worst models I've used. Every prompt felt off‑track, the answers were shallow and sometimes outright wrong, and I kept hitting bizarre, unsafe suggestions. The experience was frustrating and a huge waste of time, leaving me uneasy about relying on it for any serious work.

I tried using Gemini to work with both documents and images in a single session, but it just stopped supporting that combo. The tool threw errors and refused to process the files together, which made my workflow clunky and forced me to split tasks manually. It was irritating to hit this wall when I needed seamless multitasking.

I spent two days trying to get a nano‑banana image in a 16:9, 1400 × 728px size, but every result kept coming out square 1:1. I even used a “critical prompt” to force the ratio, yet it never changed. After hours of tweaking prompts I’m still stuck with the same wrong aspect ratio, which is pretty frustrating.

I’ve been following the hype for months, but the new 3.0 Pro feels like a step backward. The context window shrank, hallucinations spiked, and overall it performs worse than 2.5. It’s infuriating to see a supposedly upgraded model mess up basic tasks, and I’m left wondering what went wrong and if a fix is ever coming.