Here’s a use case for AI that I really like: a Japanese to English translator that gives me a translation and breakdown of the kanji (Chinese characters) in a Japanese phrase or passage. The reason AI is really good for this as opposed to Japanese apps that let you paste in phrases or do text detection in images is that you get translation, the ability to specify how you want the results returned, the image text detection, AND the ability to ask follow-up questions.
I created both a Raycast command and a Raycast AI chat preset with instructions on translation from Japanese to English.
The Raycast AI Command:
The Raycast AI Chat Preset:
The command opens in Raycast Quick AI view and uses the contents of the clipboard as the source (including screenshots or photos). The chat preset is for the separate AI chat window and I can paste or attach whatever I like as the source.
Here’s an example use case:
I took a screenshot of an easy Japanese lesson from JapanesePod101.com and gave it to the chat preset to translate.
やめてよ。「野沢」か、「菜味」って呼んでって何度も言っているでしょう。
Here’s the full translation result, including a follow-up question of mine as to why it used one specific word for “many times” instead of another.
Notification: Preset: J to E translator
---
Scott Willsey:
---
Raycast AI: Okay, here is the breakdown and translation:
**English Translation:**
Stop it! I've told you so many times to call me "Nozawa" or "Nami", haven't I?
**Japanese Transcription (Kana):**
やめてよ。「のざわ」か、「なみ」ってよんでってなんどもいっているでしょう。
(Yamete yo. "Nozawa" ka, "Nami" tte yonde tte nando mo itte iru deshou.)
**Kanji Breakdown:**
| Kanji | Reading(s) | Meaning | Compound Word(s) | Compound Reading | Compound Meaning |
| 呼 | yo(bu), ko | call, invite, breathe out | 呼んで | yonde | Call (te-form of 呼ぶ) |
| 何 | nan, nani | what, how many | 何度も | nando mo | many times, repeatedly |
| 度 | do, tabi | degrees, occurrence, time | 何度も | nando mo | many times, repeatedly |
| 言 | i(u), koto | say, word, speech, statement | 言っている | itte iru | are saying, have said |
*\*Note: 菜味 (Nami) is one possible reading for this name. Other readings exist.*
---
Scott Willsey: why 何度 intead of 何回 ?
---
Raycast AI: That's a great question! While both 何度 (nando) and 何回 (nankai) translate to "how many times" or "how often," there's a subtle difference in nuance and usage.
1.**何回 (nankai): Focus on Countable Occurrences**
* This tends to focus more on the specific, countable *number* of times an action happened or will happen. It's often used when you could potentially answer with a precise number (even if the speaker doesn't know it).
* Think of it like counting discrete events.
* Example: 「この映画、**何回**見ましたか？」 (Kono eiga, **nankai** mimashita ka?) - "How many times have you seen this movie?" (You could answer: 3 times, 5 times, etc.)
2.**何度 (nando): Focus on Repetition/Frequency (Slightly More Abstract/Emphatic)**
* This often emphasizes the *repetition* or *frequency* itself, rather than just the exact count. It can feel slightly more abstract or carry a bit more emotional weight (like surprise, exasperation, or gratitude for repeated actions).
* It's very common in the pattern **何度も (nando mo)**, meaning "many times," "again and again," or "repeatedly."
* Example: 「**何度も**すみません。」 ( **Nando mo** sumimasen.) - "Sorry [for bothering you] so many times." (Focus is on the repeated nature of the bother).
* Example: 「**何度**言ったらわかるの？」 ( **Nando** ittara wakaru no?) - "How many times do I have to tell you?" (Expresses frustration at the repetition needed).
Here, **何度も (nando mo)** is used. This emphasizes the *repeated nature* of the speaker telling the listener how to address them. It conveys a sense of "I've told you this *again and again*." The focus isn't necessarily on *exactly* how many times (was it 5 times? 10 times?), but on the fact that it has happened *repeatedly*, implying frustration.
While you *could* technically say 「何回も言っているでしょう」 (nankai mo itte iru deshou), **何度も** feels slightly more natural and common in this specific context of expressing exasperation about repeated instructions or warnings. It highlights the *frequency* and the speaker's feeling about it.
Not only does this give me a full breakdown of the kanji, including their compound and individual meanings, but it lets me ask questions about specific parts of the sentence.
Bear in mind I lived in Japan for 8 years and I’m going to know when an LLM gives me something bogus or tries to lead me astray. But this is good for reminding me how specific conjugations work or the differences in terms that I might not understand.
I was a kid when I lived in Japan, and I never formally studied Japanese. My learning was based on intuitive understanding as I played with other kids, watched TV, and spent time listening to people around me talking. Inevitably I didn’t learn some nuances or made incorrect assumptions that then became part of my understanding of Japanese. That’s why use cases like this, with a custom kanji breakdown instruction and the ability to ask questions about the translation or the source phrase, are super handy for me.
Here’s what my current Music Reviews page looks like. The goal is to start adding them in a way that creates a diverse mix of music while I build up the number of reviews available.
IndieWeb is where the fun is… if you like the random stuff I throw on my site, think about how to make your own or what you want it to be to try to make life just a little bit happier for a minute or two for anyone who stumbles across you. We could all use some good stuff in life right now.
When I added a Reviews page, I wanted to be able to have individual reviews show up in my site search. Due to some complexities I won’t go into of how Pagefind indexes things and how I optimize my images using the Astro Image component, this was easier said than done – until I remembered the good old HTML hidden global attribute. Then I just made a hidden span and populated it with the image alt text.
I’ve spoken a couple times about my Sessions Raycast script command, both when I originally set it up, and then again when I updated it to use Moom as my window manager. Well, I’m back to using Raycast as my window manager, and I also use a Raycast focus mode for my podcast session. So here’s your updated Session Raycast script command, 2025 edition.
You may recall that the purpose of this script command is to just set up my environment in a way that’s suitable for a specific task. It always closes all apps, then opens some core apps, and then opens other apps specific to the session type.1 It also sets audio output and input based on the session type, arranges app windows in a specific way, and either sets Coffee for Raycast to Caffinate mode (keep the screen on) or Decaffinate mode (let it sleep when it wants).
When I run the Sessions script command in Raycast, I just choose the session type. The nice thing about Raycast dropdowns is they pre-select the option you chose last time. In the case of the Session script command, that’s not as convenient as it is for some other script commands, but that’s life.
Anyway, if you have any questions about this script command or Raycast script commands in general, you know how to find me (hint: Mastodon or Bluesky).
Footnotes
I should see if there’s a way to close all apps that aren’t in a list, instead of closing all apps and then re-opening all apps that are in a list. ↩
Last week or so, I started writing a blog post as I sometimes do, this one pertaining to my Automation Workflow for Media Reviews. Unfortunately, I wanted to preview it as I went along, so I copied it in progress to my git main branch of the local copy of the website.
You can see where this is going.
Yes, I updated something else on the website and published it, INCLUDING the partial draft of the blog post I was working on. This wasn’t a super huge deal, except that I use EchoFeed to automatically post to Bluesky and Mastodon whenever I post something new on the site.
Sigh.
The good news is that this finally pushed me to add drafts functionality to my site, so that I could have drafts render when running locally in development mode, but not actually get written when doing a site build. It’s a good, basic feature to have.
I’ve always wanted to be a writer, and I’ve always wanted to create drafts in Astro that won’t get published until I want them to.
10
11
[Now I can!](https://jacurtis.com/notes/astro-draft-posts/)
One more thing though – none of this keeps the post page itself from being rendered during a build. It just keeps anything from linking to it or showing it in a list of posts. This means that it will show up in your RSS feed unless you edit your RSS template to also filter it out.
The now page is a cool concept by Derek Sivers for talking about what you’re doing now in your life. I use my /now page slightly differently, to just list things I’ve been reading, drinking, eating, watching, listening to, and sometimes playing recently.
Initially everything on my now page was bullet lists of text links. This works, and it’s fine, but I also was a little envious of how nice some other people’s media logs looked with artwork for their movies and shows. I’m thinking of Robb Knight’s Almanac, Habib Cham’s Books page, and Cory Dransfeldt’s Music, Watching, and Currently Reading pages, for example.
As I was rolling how to improve my /now page around in the back of my mind, Matt Birchler release a neat little iOS app called Quick Reviews. Quick Reviews is based on a web utility he’s had for awhile now which lets you create little images with media reviews, complete with artwork, and share them with people.
As I played with the iOS app version of Quick Reviews and saw how it exported the images, I realized I could build a workflow around them for incorporating them into my /now page.1 Most importantly for any change to the page to make it look better was that it not take much more time to update on a regular basis. The text link version was easy – I have scripts that pull my Trakt history for the Movies and TV Shows portions of the page and return formatted markdown bullet list items. All I have to do is paste them where I want them.
Thanks to the power of scripting and the wonderful array of automation utilities for macOS, I’ve been able to find a way to make updating my /now page with Quick Reviews images equally quick and painless. So get ready, because the rest of this article exists to show you how it’s done.
By the way, I made a handy little video of the automation workflow in action which you can watch here or on YouTube, or jump to the detailed workflow description below to keep reading.
Automating Quick Reviews
Step 1: Quick Reviews
The first link in my automation chain is Quick Reviews by Matt Birchler. This is an iOS app that lets you choose a piece of media to review and allows you to add artwork, your rating, and your text review to a nicely formatted image.
First you choose a Media Type (Movie, TV Show, Game, Book, Music, or Other), give it the title of the media, and either add artwork or use Magic Mode to add one if it’s a Movie, TV Show, or Game, and then add your text review and choose a rating. You can customize the text that gets used for ratings to your own liking (⭐⭐⭐⭐⭐ use emoji stars if you want!) at an app level, by the way.
Next, edit the image colors for background and rating highlight, edit the font and font styles, and save or export the image. In my case, I use export and share and then AirDrop it to my Mac.
Step 2: Hazel
When I AirDrop the Quick Reviews image to my Mac, several things happen automatically, thanks in large part to a wonderful Mac utility called Hazel, by Noodlesoft. Hazel does one thing: it watches folders that you tell it to. But then it allows you to perform a ton of different actions on those folders, or files in those folders, based on triggers you specifiy.
In my case, I have Hazel watch my Downloads folder for incoming images that have the png extension and that have file names containing “PNG image”. These criteria are based on the exported Quick Reviews file name and type. If any files meeting those criteria hit the Downloads folder, they’re moved to /Users/scott/Scripts/scottwillsey/scottwillsey-images/In, renamed in a specific way, and then a Bash script called scottwillsey-media-review-images.sh is run.
Here’s the entirety of that script:
scottwillsey-media-review-images.sh
#!/bin/bash
# Required parameters:
# @raycast.schemaVersion 1
# @raycast.title ScottWillsey Media Review Images
# @raycast.mode silent
# Optional parameters:
# @raycast.icon ../icons/scottwillsey.png
# @raycast.packageName Website
# Documentation:
# @raycast.description Resizes and copies images for the ScottWillsey site's /now page media reviews.
You can see that it’s written as a Raycast script command, which is why all the weird comments are in there, but I don’t run it from Raycast because I let Hazel call it directly when I AirDrop the image.
First, it runs a Mac app called Retrobatch with a specific Retrobatch workflow, waits for that to finish, and then runs a Python script.
Which takes us to Step 3…
Step 3: Retrobatch
Retrobatch is a stupendously amazing image automation tool from Flying Meat Software. Its whole job is image batch processing.
The particular Retrobatch workflow that I run in this case takes images from /Users/scott/Scripts/scottwillsey/scottwillsey-images/In, where Hazel moved them, and creates two copies – one resized to 1000px wide and saved as a png, and the other full-sized but saved as a jpg. Both of these are saved to /Users/scott/Scripts/scottwillsey/scottwillsey-images/Out.
The purpose of having two copies is so that I can display an optimized version in my articles and have those link to the full-sized image if the reader clicks on them – more on this later.
One other important thing this Retrobatch workflow does is to extract the text from the resized image and save that text in a file in the Out folder with the same file name as the corresponding image, except with a .txt extension.
I need the text from the image to create my image link alt text for the website, and the fact that Retrobatch can extract this for me makes the fully automated nature of this process possible. I don’t have to make a copy of the text while typing my review on the phone and send it to the Mac somehow. It’s literally delivered in the image and extracted on the Mac by Retrobatch. It’s beautiful.
Now Retrobatch is done. For each review image in /Users/scott/Scripts/scottwillsey/scottwillsey-images/In, there is now a corresponding resized png, full-sized jpg, and text file with the text from the image saved in /Users/scott/Scripts/scottwillsey/scottwillsey-images/Out.
When Retrobatch finishes, the scottwillsey-media-review-images.sh script that started the Retrobatch workflow moves on to the next step – running a Python script called scottwillsey_review_images.py.
Step 4: Python Script
The Python script is much more involved than the Bash script was. I’m just going to show you the whole thing and then walk though it.
This creates a UUID to add into the image file names. Then it looks at every file in the Retrobatch output directory (/Users/scott/Scripts/scottwillsey/scottwillsey-images/Out), checks to see if it’s a png, and gets its base file name if so.
After getting the png file base name, the script looks for a corresponding txt file, which will contain the extracted review text. If it finds one, it reads the contents of the file to use as image alt text.
Next, it creates a markdown link for the png image complete with alt text, and a link to the larger jpg image. Then it appends this markdown link to the file_links variable.
Finally, it copies the complete file_links variable to the clipboard. This results in markdown for the images ready to paste into the /now page. The markdown basically looks like this, albeit with a few extra bits of text from the extracted text that I clean up:
[](/images/posts/PngImage42999A84080-review-6126286a-a138-4551-8892-ad278b0bfa5f.jpg)
Python Script Part 2: Archive All Images in the Retrobatch Input Folder
# Copy files to site public images directory and move files from input to review images directory
forfilein os.listdir(input_path):
# Get base name and extension
base_name, ext = os.path.splitext(file)
# Create new filename with UUID
new_filename =f"{base_name}-{new_uuid}{ext}"
src_file = os.path.join(input_path,file)
# Move file to review images directory
dst_file = os.path.join(review_images_path,file)
shutil.move(src_file, dst_file)
This part is simple. It just renames the images with the UUID included in the file name and moves them from the Retrobatch input folder to a review images archive folder. This is just a folder for me to keep copies of my reviews in, it has nothing to do with generating the website.
Python Script Part 3: Delete All Text Files in the Retrobatch Output Folder
# Delete all .txt files in output directory
forfilein os.listdir(output_path):
iffile.endswith(".txt"):
os.remove(os.path.join(output_path,file))
Again, very simple. This just deletes all the text files in the Retrobatch output folder. I don’t need them anymore as I have the text in the markdown links.
Python Script Part 4: Rename and Move the Image Files to Use on the Website
# Move files from output to site src assets images directory
This loops through all files in the Retrobatch output folder (since only images remain here) and does the following with each:
Gets the base file name and file extension,
Creates a new filename with the UUID added in between the base file name and extension,
Moves the image to /Users/scott/Sites/scottwillsey/src/assets/images/posts with the new UUID-included file name if it’s a png file,
Or moves the image to /Users/scott/Sites/scottwillsey/public/images/posts with the new UUID-included file name if it’s a jpg.
That’s the entirety of the script. Once it’s done, there are no images left in the Retrobatch input and output folders, there are jpg and png images in the correct website source folders for use on the compiled /now page, and there are corresponding Markdown links in the clipboard ready to paste into the /now page content source Markdown file.
Step 5: Profit
The results are something like the image below, or actually what you see if you click the link for the /now page on the site menu. It’s the one that looks like a clock.
There are definitely improvements that could be made, but this is a great start, and it’s proof that sometimes things that look like a very manual process can be automated, thanks largely to the great ecosystem of Mac indie programmers who make outstanding irreplaceable utilities like Hazel and Retrobatch.
If you want to see a comparison of the old style, imageless /now page and the current beautiful one, just look at the image below. On the right is the new, Quick Reviews-ified version, on the left is the old one. Or, to put it another way, on the left is the before now, and on the right is the now now.
Summarium
That’s a lot of words, so let’s summarize the process here:
Scott Willsey creates a review in Quick Reviews on iOS,
Scott Willsey AirDrops it to his Mac,
Hazel renames and moves the image and runs a Bash script,
Bash script runs a Retrobatch workflow and waits for it to finish,
Retrobatch makes two copies of the image, a full-sized jpg and a smaller png, as well as a text file containing the extracted text from the image,
Bash script runs a Python script,
Python script constructs the Markdown links and moves the Retrobatch images into the correct locations for the site to compile and display on the /now page,
Scott Willsey pastes the Markdown links into the /now page source Markdown file,
Scott Willsey says “Damn, that is cool!” and goes to make another cup of coffee.
Footnotes
If you’re asking yourself right now “Is this guy going to make EVERY reference to his /now page a hyperlink to it?”, the answer is yes. Yes, I am. ↩
You may have noticed that the site menu has a couple new icons in it. One looks like a timer or stopwatch – that’s a link to the /now page, which I’ve had for awhile, but haven’t had a menu link to. The other looks like movie director clapper with a gear on it – that’s a link to a new page: my /reviews page.
You may look at both the /now page and the /reviews page and think “yeah, but they’ve both got these little reviews on them”, and that’s true, but there is a difference. The /now page reviews are the latest only, and some of them (TV series and books, for example) may be things still in progress. These are ephemeral – they’ll drop off as I move on to new stuff.
The reviews page is only items I’ve finished, and is ALL of the items I’ve finished that I’ve written reviews for. Nothing will drop off here. Eventually, then, I’ll probably need to break each category (Movies, TV Series, Books, Games) off into their own pages so it doesn’t take an hour and a half to load the page. 45 minutes I could live with, 90 minutes not so much.
I’m working on an article with an accompanying video about my workflow for creating and publishing these reviews. I think you’ll find it interesting to see how much of the process I’ve managed to automate.
