+1 for Joplin - but unless something's changed in the past few years, I can't stand the way they structure your notes.
If you already have a well organized folder/subfolder of markdown notes, Joplin takes'em and shoves them into a SQLite table with UUID names, attached media is also renamed and stored under something like `~/.config/joplin-desktop/resources`.
Makes it very difficult to use your existing notes with bash tools, etc. once Joplin has ingested them.
Fenced gardens are a -5 for any note taking utility. If the tool can't/won't use a filesystem to organize documents as files in folders, I assume the author was a moron. I don't mind if it has its own structured storage model as well for whatever reason, but skipping out on the most foundational persistence structure is an instant fail, being (amongst other sins) a mismatch for work trees. And import/export is friction, not an exemption.
I have a second-hand Samsung Frame connected to an old laptop that periodically wakes up, generates a couple dozen images daily, and then transfers them to the Frame TV via the Samsung Frame’s WebSocket API. It automatically changes the picture every 30 minutes.
The "Art Mode" setting along with its matte finish makes it non-reflective, so it just acts like a nice little gallery on my wall. It’s using older models, mainly SDXL, so a lot of the pictures can be a little goofy and its been humming along for several years at this point.
I’d love to pick up one of these TRMNL devices and do something similar but the price at $1200 (soon to be $1300) gives me pause.
Not the GP, but I did a similar thing where an eink picture frame displays AI-generated landscapes with the weather being the same as the current weather/time of day as outside.
I just like the variation and didn't need to go source a bunch of actual images of the specific weather patterns.
Thanks for this, I too have an older Frame TV in my living room. Want to explore exporting my calendar and photos to it from a home server via that API!
TRMNL is a cool product, I have the original. This new model looks exactly like what I wanted but certainly not in the budget. Now I have another option!
I purchased a Frame (43”) a few years back. For the specific purpose of being a TV with an Art Mode that plausibly blends in to my living room, it succeeds. Although I do ding it for charging a yearly $75 subscription for access to the updated curated catalogue of art. A few points back again for that sub actually being, in a way, almost worth it. The art is good, is what I’m saying.
Which is a hell of a thing because the smart tv software is so aggressively hostile and awful that it makes me want to give up on using it. The worst experience from moment one on. You can plug it in to an Apple TV and get mostly there but then you need to fight with the remote to get it to reliably switch to art mode.
I do not recommend it, but it’s possible that with some hacking it could be good.
The screen quality is lovely and very good at displaying artwork. It doesn’t look like a TV screen; not exactly like a picture either, but it’s really quite impressive.
The software is appalling in so many ways. Aggressively user-hostile except when it’s upselling at you.
Honestly, I would have the same reservations. Mine's an older model, but Samsung has only become more user-hostile as time has gone on.
That being said, I was able to pretty easily connect my Samsung Frame to the LAN and then block it at the router level from making outbound requests to the internet.
Mine is an old 2022 model so its not OLED - I measured it a while ago (43") and in art mode it was drawing down around ~35-40W on average about 5-6 bucks a month when left on continuously.
Back in the day running something like Super Mario 3 often meant composite video going through RF hookup and the TV "recovering" and decoding. I'm sure that accounted for a bit of the blurriness.
Contrast this with a game like Commander Keen for DOS running in 16-color graphics on a VGA CRT monitor which sends separate R, G, B signals. I remember Keen being very crisp even as a kid who practically pressed their face against the glass when playing.
I mention this a lot: the two basic post-processing steps you have to do, at a minimum, to give genai images even a semblance of pixel art are palette reduction and square grid alignment.
Unfortunately, the definition of "pixel art" like "vibe coding" has been diluted to the point of almost being meaningless.
If you have a project like this with tens of thousands of tiles, you have to run them all through square grid alignment. That process tries to pick an optimal square grid and then reduce the image to a discrete set of pixels, with each square resolved to a single color.
It's similar to what you do when you shrink an image using nearest-neighbor scaling. As a result, you mitigate a lot of the irritating high-frequency noise and fringing that occur in these higher-resolution pixel art images from generative AIs.
Applying it to a project with tens of thousands of images still requires a lot of manual intervention. This is what makes separates good pixel art from bad pixel art: even at a low resolution of 16x16 or 32x32, it's still recognizable.
Try taking any of the 512x512 tiles from this particular isometric city project (or any of the other dozen variants) and feed'em into Spritefusion and you'll quickly see what I mean.
> Unfortunately, the definition of "pixel art" has been diluted to the point of almost being meaningless.
Nah, people who enjoy pixel art games, and pixel-artists have complete clarity on what pixel art means. There is zero dilution to that term among people who actually have pixel art drawing talent.
Did you ever play Pac-Man Vs.? It's an old game released in the early 2000s by Nintendo where one player takes control of Pac-Man and the other players control the ghosts trying to catch him. It's a lot of fun.
KQ 5/6 are two of my favorite games of all time. To this day, whenever I come across a snake while hiking, I immediately channel my inner Cedric to the annoyance of every other person in my group.
Right, you can be an incredibly articulate speaker without necessarily knowing the grammatical terms for the different parts of speech in each sentence.
I think there's a number of people who learned to play by ear who just sort of "internalized" theory without needing a formal curriculum.
So I've done a little bit of testing. When I loaded up the editor, I expected it to communicate with the installed binaries for my large language models, or possibly use a BYOK (bring your own key) approach. But once I understood that it's driven from the command line and you see the changes happen in the editor that gets spun up, I had no problems.
I've been experimenting a fair amount with LLMs driving procedural graphics myself, although my experiments have been mostly limited to relatively low-resolution stuff, such as EGA adventure game graphics.
So I decided to put Klad through a relatively hard test: I took a reference picture of an illustrated squirrel and asked it to recreate it, then set up the rigging for a basic walk cycle. This was using Fable 5.2 high.
Then I did the same thing, but with a more traditional approach, using standard generative models to manually create a walk cycle. This was all done relatively quickly, so bear that in mind.
Here are the results. I still think these are early days, and LLM-driven graphics are only going to improve.
And there are obvious advantages to a procedural system with rigging: Much easier to maintain consistency, palette control, and the ability to make precise alterations without having to deal with any kind of conventional in-painting nonsense.
Hey thanks so much for putting Klad and Fable through this test!
I think the results are really interesting. I think it's very obvious the classic image-to-video approach yields much more natural results at this time and with the current models.
I think I'm going to do some testing of my own, and see where things can be improved. I do think that the bottleneck at this time, is the model you use itself.
But it's like you say, it's very early days with an LLM expressing its ability to draw/paint. In my own testing, it kind of feels like it did with coding 2 years ago: yeah sure the model is pretty Okay at building a function, but I wouldn't have let it build a whole feature back then.
Nowadays, I'm practically only reading code and thinking about how to refine my instructions for the next build.
I'm sort of betting that this evolution could go the same way for graphics (and possibly audio/midi as well). And I wanted to have a tool ready, to see how it evolves.
> I do think that the bottleneck at this time, is the model you use itself.
I definitely agree. For me, using an LLM to drive drawing tools for procedural graphics only really became viable in the last year with the advent of open-weight model Qwen3.8-27B and proprietary models like Opus 5x.
If you’re interested, I built something in a similar vein: a custom drawing language that renders old adventure game-style scenes and is also driven by an LLM.
It produces relatively decent results because the domain is inherently constrained to an EGA color palette and a low resolution designed to approximate old adventure games from the 1980s.
And like your tool - half the fun is watching the LLM drive the production of the final image in real time!
> An image model trained on a hand-picked pixel-art style then repaints it, tile by tile, each tile continuing its painted neighbours so the city stays seamless.
Nice work. The developer of Isometric NYC did something similar - creating a fine-tune of the Qwen Image Edit model. I'd be curious to hear more detail about the image model you used. (the base checkpoint, LoRAs, etc).
If you already have a well organized folder/subfolder of markdown notes, Joplin takes'em and shoves them into a SQLite table with UUID names, attached media is also renamed and stored under something like `~/.config/joplin-desktop/resources`.
Makes it very difficult to use your existing notes with bash tools, etc. once Joplin has ingested them.
reply