Proof 1: AIStudio Was Nearly Unlimited
Believe it or not, Gemini 2.5 Pro was the undisputed king through early-to-mid 2025. It’s not like I haven’t tasted the good stuff — I ran a GPT5 subscription for a while, Thinking model included, plus that Pro model it bragged was so mighty — but in actual output latency and willingness to speak human, neither matched the 2.5 Pro I used for free on AIStudio.
2.5 Pro was also the first model that felt stronger than me personally, one that could actually take some heavy lifting off my hands. And every one of my exam reviews ran assisted by this prompt; without an efficient model, the efficiency would surely have taken a beating:
现在我正在进行这门科目的学习，你的要求如下，必须严格遵守：1.对于日常语言，你必须使用中文，但是如果遇到你认为必要或者关键的英语专业名词，可以适当使用英文表示或者双语对照（但是，如果专业名词或者关键词没有合适中文，可以不用双语），这不仅在这次解释中要遵循这个规则，在以后的对话中也要遵循。2.根据课件内的内容，首先大概的过一遍课件内所囊括的知识点，这一边是为了让我熟悉课件内容，并带有讲解。然后，仔细的讲每一页课件中所有知识点与细节(NOTICE: PAGE BY PAGE)，如果必要，附上例子（可以为知识点应用例子，也可以是代码块）。3.有些知识点稍微复杂和难理解，你可以在这之外适当加入一些容易理解的内容，但注意这种内容当且仅当你认为这个知识点很难理解。4.课件内无关的课程信息不需要讲解。然后，在最后，给出一些面向关键概念或者容易混淆、不易理解知识点的小测试（quiz）并附上解答。5.禁止在你的回答中加上cite或者[cite_start]，你只用线性过一遍课件。6.同时记住这个提示词，我们的对话会包括多次课件内容解释以及答疑解惑，在我提出的问题中你要积极使用课件知识，并且在我传入新的课件内容的时候要继续严格遵循以上规则。同时，如果你可以设置这次对话的标题，将其设置为 例如[CS203(if exist), or Compiler(if not found)][lecture number] Overview。
(“My courseware-cramming prompt: stick to Chinese with English terms where needed, walk through the slides page by page with explanations and examples, skip unrelated course info, and finish with a quiz on the tricky concepts.”)
Sadly, AIStudio has since been gutted — daily usage has fallen to nearly unusable levels (rumor has it 10–20 turns per model). But that doesn’t stop us. Behold, Proof 2.
Proof 2: Google One One-Year Student Trial
13 kuai won’t even buy you a 1+1 at McDonald’s, but on Xianyu it can get SheerID verification and card binding sorted. That means a one-year student-trial Pro package: Antigravity included (these past two days I’ve been trying to reverse-API my way into CC, but my account region keeps acting up), nearly inexhaustible usage of the Gemini website, and 2TB of cloud storage.
Honestly, this maneuver does carry a whiff of fraud — but am I not a college student? (flees)
Proof 3: Nano Banana Pro
I haven’t used many image-generation models; word is the market has plenty of strong ones, but most of them charge. I used to run Nano Banana off a one-month enterprise trial. With the Google One above, I now hold a full year of unlimited Nano Banana Pro rights.
What do you mean, you don’t know what this thing is for? See the examples below:
Nano Banana Pro nailing it on the first draft
Nano Banana Pro after a few attempts, plus some manual layout edits
Nano Banana Pro on a complex scene: still flawed after multiple drafts, but barely presentable
At my current stage of study, Nano Banana Pro is fully up to the figure demands of reports. One simple image-generation prompt multiplies your report’s polish several times over — a genuinely can’t-lose business.
And do you really need to write the prompt yourself? As it turns out, one universal meta-prompt is enough:
帮我把这一块的内容转化成为一段用于生成图像的prompt。我们的图像风格应该是严谨的IEEE Trans插图风格，应该看起来简洁且干练。你需要先构思如何巧妙的转换成一个容易理解的配图，然后仔细的描述图片的每一个部分应该有哪些元素，元素之间的位置关系，以及硬编码一样的显式的写明在哪里应该写哪些文字。
(“Universal meta-prompt: turn this section into an image-generation prompt — strict IEEE Trans figure style, clean and crisp; plan the figure first, then describe every element, its placement, and hard-code exactly which text goes where.”)
The rough flowchart looks like this — and how this image was made, no need to retell:
Well. Having freeloaded this much off Google, I do feel a bit ashamed. If someday I get the chance to be exploited back by Google, maybe that would make me feel less ashamed. (Daydreaming about eating and taking at the same time…)