Ashita

index

Blog

blog

Works

works

About

about

With AI helping me untangle a 2017 Stack Overflow conversation, I finally understood the reply—and remembered the translation struggles, downvotes, and person who helped rewrite my question.

Kindness from Nearly Ten Years Ago

stackoverflow-kindness

After trying Astra, I used Doro and the goofy cats I like to explore unfamiliar 3D work, and tried video generation too. Later, I posted on Xiaohongshu and got roughly 3k likes. The notifications made me happy.

I Tried AI Modeling and Became a Xiaohongshu Creator for a Few Days

doro-3d

I wrote Vido by hand with Vue 1 in the summer of 2016. Ten years later, version 2.0 is playing videos in my blog, with its own colors and a little avatar.

A Video Player, Ten Years Later

vido-ten-years

Koe grew out of mimi as an attempt at general-purpose offline video captions. Which part of a long video to process, how to get its audio, and which player the captions belong to turned out to be difficult questions.

Koe: the browser subtitle extension that didn’t work out

I first added ten yuan to rent a GPU and set up the software, then tried hosted workflows on AutoDL.Art. These are the costs and waits from both attempts.

Two Ways I Tried Making AI Video on AutoDL

autodl-h3

Find the text, recognize the Japanese, translate it, and put it back. Column breaks and painted-over backgrounds can still make the page awkward to read.

Putting AI-Translated Text Back into Manga Bubbles Wasn’t So Simple

manga-ocr

Changing the vocal sound of an existing performance with RVC: separating vocals, preparing training data, extracting features, training a model, and building a retrieval index. I got everything ready, then stopped during training.

How an AI Cover Changes the Voice

ai-cover-from-beginner-to-giving-up

Around H3’s launch, a transformation video on Bilibili made me try one with Gwen. Six splash illustrations became five transitions, joined and given sound afterward, for a finished clip of about 26 seconds.

Trying MiniMax H3 Video Generation for the First Time

big-sucker-ai-research

mimi turns system audio into live captions and translations. Words can change after appearing, but waiting for them to settle can miss the scene. I’m still working on that trade-off.

Subtitles for whatever my computer is playing

mimi

At a company AI hackathon, we combined video frames into a grid. It helped with debugging, but sampling could skip an order number that appeared only for a moment.

Putting video frames together for AI

ding-frame

A small server and the sites my wife and I use led me to build nido. Frontend releases and SPA deployment grew into version history, shared scripts, pages, and site management.

nido: Frontend Infrastructure on a Small Server

nido

AI helped me make projects I had kept putting off. Once the code runs, I still want to know whether it does the right thing. I don’t know whether checking it will always need me.

Can I Trust the Code AI Writes?

ai-era-evolution

My iPhone 12 still had trade-in value after five years. The service wouldn’t take my high-end MacBook once it stopped turning on. Wanting a replacement doesn’t have to mean waiting for the old device to break.

My Old Phone Had Trade-In Value. My MacBook Was Rejected.

digital-minimalism

In January I connected an iPhone Shortcut to my ledger, X, to read a payment screenshot and let me confirm the entry.

Logging Expenses from Payment Screenshots

x-bookkeeping

There is nothing particularly wrong with my life now. I just rarely start anything on my own. Writing Ashita again is one small attempt to change that.

[tag]

# Experiments

I Tried AI Modeling and Became a Xiaohongshu Creator for a Few Days

Two Ways I Tried Making AI Video on AutoDL

How an AI Cover Changes the Voice

Trying MiniMax H3 Video Generation for the First Time