Gemini felt dumb on December 21, 2025.

I tried to download the full‑size Nano Banana Pro images from Gemini, but every click gave me the same old picture from earlier in the chat instead of the new one I’d just generated. It forced me to settle for the low‑resolution JPGs and I ended up losing a bunch of iterations. The whole thing was infuriating and made me worry about missing important work.

I’ve been running a long‑term chat with Gemini and suddenly the AI’s replies vanished, even though my prompts are still there. Uploaded files now show “file has been removed,” which is super annoying because the conversation spanned a model update and was carefully curated. I’ve started exporting incrementally to Markdown and re‑uploading to keep things alive, but it feels like a major limitation for lengthy sessions. I’m not sure if this is a bug or a setting issue, even though my session retention is set to keep everything. Any insight would be great—maybe I just need to back up more often.

I’ve been running Gemini 3.0 Flash a lot since it launched, and while the reasoning and prompt‑following feel sharper than 2.5, the spelling is a nightmare. Every few thousand words I spot 4‑5 glaring typos—“Envirnoment,” “Repsponse,” “Integegration”—as if the tokenizer broke. It makes long‑form writing feel pointless unless I run a spell‑check afterward, which defeats the purpose of a supposedly SOTA model.

I keep seeing Gemini pop up follow‑up questions in bold text, and it’s driving me crazy. It feels forced and annoying, like the tool is pushing me into extra steps I don’t need. I’m looking for a way to stop it because the constant bold prompts are really irritating.

I experimented with the two Gemini modes and noticed a clear split: the Pro setting gave me concise, spot‑on answers that felt reliable, while the Thinking mode added a quirky, chatty personality that made the responses feel more human but also a bit rambling. I liked having the choice—when I needed precision I switched to Pro, and when I wanted a more lively interaction I went with Thinking. The contrast was striking and helped me decide which vibe suited each task.

I keep getting the same vague refusal message when I ask Nano Banana to edit AI‑generated characters, even when the images aren’t based on any real person. The app says it can’t edit certain public figures, which makes no sense for my prompts. It feels like the tool is blocked or mis‑configured, and I’m left wondering if it’s an Android app bug or something else.

I tried the same prompt on ChatGPT and Gemini 3.0 Pro, expecting thorough reasoning and source citations. ChatGPT took almost 13 minutes and gave linked references, but Gemini buzzed through in 20 seconds, left out any citations, and produced a clearly incorrect answer. It happens with every prompt, and I can’t coax it to think longer or cite sources, leaving me frustrated and wondering if I’m using it wrong.

I’ve been using Veo 3.1 for ages, and after finally trying Sora 2 I could instantly feel the jump in realism and physics. The new model feels way more lifelike, and I’m blown away by how smooth the motion looks. It makes me wonder when Google will finally bring that same magic to video generation – I’m excited but also a bit impatient for the next breakthrough.

I love Gemini’s personality and its solid answers, but the random follow‑up questions are driving me nuts. After getting a spot‑on reply about cricket, it tossed in a completely unrelated query about my diet and medication. It feels like the model is defaulting to unnecessary chit‑chat, and I’m left wondering if there’s a way to turn that off or if it’s just hard‑coded. This annoyance really dampens the otherwise good experience.

I tried using the Nano Nanana Pro’s image creation tool, picking the pro model and sending a prompt that should return an image. Instead, it kept spitting out regular text like Gemini 3 Pro. I even tweaked the prompt to explicitly request an image, but the tool still gave me text. Every time it failed, I had to start a new chat, which was incredibly annoying and broke my workflow. This inconsistency is really frustrating.

I’ve been enjoying Gemini’s subscription for weeks, but over the past 3‑4 days the Fast and Thinking modes just quit on me, spitting errors on every request. The Pro mode still works, and the desktop web version is fine, but on the iOS app it’s a dead end. It’s really frustrating to lose those modes when I need them.