Claude and I have spent most of the summer on exactly this problem for Skechworks (skechworks.com), a free little Mac design app I made to help my own workflow that is very similar to yours. I use it for my custom coin shop The Achieve Mint (theachievemint.com).
It runs VTracer under the hood too. Most of the cleanup you're doing in Inkscape could possibly be pushed to before and after the trace?
Skechworks picks the palette before the trace and forces it again after. I quantize to 8 colors with pngquant, trace, then snap every fill back to the nearest palette color. VTracer fills each region with its own average color, otherwise you get hundreds of nearly identical grays. One design went from 279 fill colors to 3.
I upscale 2x before I quantize. The anti-aliasing is what tells the tracer where an edge really sits. Quantize at 1x and you throw that away, and small text comes out lumpy. Related gotcha: filter_speckle is an area, so don't double it when you double the image.
I also flatten transparency onto white first. VTracer counts every alpha value as a different color. I had a 5 color picture that it saw as 688 colors, and the trace took 6 GB of memory. Flattened, the same picture took 300 MB.
For the background I sample the whole edge of the picture. If 85% of it is one color, I delete every path in that color that touches the edge. White inside the drawing never touches the edge, so it stays.
For black and white I threshold with Otsu, then trace in color mode with hierarchical stacked instead of bw mode. bw mode hands you one giant black path full of holes. Stacked gives you every region as its own shape, which is much easier to edit. You have to pad the image with the background color first, or VTracer decides your subject is the canvas.
Your FLUX step is the same trick I use for photos and messy art. I have an image model redraw the picture as flat art, then trace the redraw. That came out about a quarter the file size of the best direct trace I could get. The catch is that the model moves and resizes the subject a little every run. I measure the box around the ink in both pictures and fit the trace back onto the original.
What I haven't solved is curve fitting. vectorizer.ai gives you real arcs and straight lines. VTracer gives you splines, so circles are never quite circles.
Something I discovered recently: if you stop following everyone you’re connected to, Linkedin’s feed breaks so you get this by default. I can still interact with everyone normally, send messages, etc, but I do not see any Linkedin content, ads, etc. Do not miss it.
That used to be the case, and I loved the empty feed. But, i noticed that it started showing me the default feed of all the AI slop people post which linkedin thinks is relevant to my field. The current way that worked for me was to select ‘Chinese - Simplified’ as the language and since I don’t understand it, I am not remotely curious to read the cringe. I now Only see if there is any messages i receive (which aren’t translated). Basically, in and out of the website in 15 seconds if no messages/connect request! Works like a charm for now.
I still do not understand how models can generate absolutely stunning SVG-like bitmaps of a pelican on a bicycle, but fail to do so when asked to directly create an SVG.
The image linked below was generated as a bitmap by Gemini and then manually converted to an SVG. Why can models not even remotely output something like that as SVG?
Try looking at the SVG code for one of those converted from bitmaps, and I think you'll understand why even a massive LLM has trouble generating it on the fly (the pelican SVGs are always one-shot single API calls, so no iterations and verifications).
This one's going to be out of left field, but last Thursday I launched Countdown Treasure (https://countdowntreasure.com)
It's a real life treasure hunt in the Blue Ridge Mountains with a current total prize of $31,200+ in gold coins and a growing side pot.
I modeled it off of last year's Project Skydrop (https://projectskydrop.com) which was in the Boston area.
* Shrinking search area (today, Day 5, it will be 160 miles, on Day 21 it'll be just 1 foot wide)
* 24/7 webcam trained on the jar of gold coins sitting on the forest floor just off a public hiking trail
* Premium upgrades ($10 from each upgrade goes towards the side pot) for aerial photos above the treasure and access to a private online community (and you get your daily clues earlier)
* $2 from each upgrade goes towards the goal of raising $20k for continued Hurricane Helene relief
So far the side pot is $6k and climbing.
It's been such a fun project to work on, but also a lot of work. Tons of moving parts and checking twice and three times to make sure you've scrubbed all the EXIF data, etc.
Thanks! Yeah, I unfortunately can't shell out a straight $20k donation, but I saved the $25k over the year from my e-commerce business and justified it as a marketing expense if worse case scenario happens and it's found earlier than the math predicts. But if we get past break even then can't wait to write that check to help out the communities around here that are still recovering.
Maybe some? But the circle today will be 160 miles wide or about the same width as Switzerland and Denmark. So I'm not sure how much shadows would help you pinpoint a specific location in an entire country worth of mountains.
I'd recommend folks watch the video – it's fascinating.
The truck gets OTA updates through your phone and not some LTE modem. It doesn't have one. They moved all car management including OBD-like functionality to the phone, too, which I think is awesome.
This is how I want the interior design philosophy of manual controls to be digitized – with digital control. I'd pay $10k more for physical buttons, though.
Only if the phone app is open source, or at least the api, alllll of it, is public so no one needs the default app nor is limited by it.
Alternatively, maybe the overall simplicity will mean that a 3rd party full computer replacement would be feasible even without any official help from the manufacturer.
I mean yes, but also this is a complex new prototype vehicle. I can assume that there may be mistakes/ non ideal things that they only catch post production.
As long as the fixes are a long the lines as bios updates (not required per say, but may fix bugs or edge cases) then that seems reasonable.
Oh sweet. Delicious. Very reassuring. Was really hoping this thing was going to be device agnostic.
My 2015 car had 3g "smart" features that no longer work since 3g has been sunset in the US. Awesome to see forward thinking of a smart feature-set that can be updated with a module you'll likley already have an upgrade path for.
There's a difference between phone-dependent and phone-augmented. I don't know the details of the truck, but I think a happy medium would be for an app to exist to augment the truck's abilities and allow at-home updates, but to not require the app or phone to just use the truck (even for long periods; i.e. you could go forever without using the app and the truck will just keep working in its current state).
I think this is technically "image variations" and I think image variations are still only dall-e 3 for now (best I could tell earlier today from the API)
It runs VTracer under the hood too. Most of the cleanup you're doing in Inkscape could possibly be pushed to before and after the trace?
Skechworks picks the palette before the trace and forces it again after. I quantize to 8 colors with pngquant, trace, then snap every fill back to the nearest palette color. VTracer fills each region with its own average color, otherwise you get hundreds of nearly identical grays. One design went from 279 fill colors to 3.
I upscale 2x before I quantize. The anti-aliasing is what tells the tracer where an edge really sits. Quantize at 1x and you throw that away, and small text comes out lumpy. Related gotcha: filter_speckle is an area, so don't double it when you double the image.
I also flatten transparency onto white first. VTracer counts every alpha value as a different color. I had a 5 color picture that it saw as 688 colors, and the trace took 6 GB of memory. Flattened, the same picture took 300 MB.
For the background I sample the whole edge of the picture. If 85% of it is one color, I delete every path in that color that touches the edge. White inside the drawing never touches the edge, so it stays.
For black and white I threshold with Otsu, then trace in color mode with hierarchical stacked instead of bw mode. bw mode hands you one giant black path full of holes. Stacked gives you every region as its own shape, which is much easier to edit. You have to pad the image with the background color first, or VTracer decides your subject is the canvas.
Your FLUX step is the same trick I use for photos and messy art. I have an image model redraw the picture as flat art, then trace the redraw. That came out about a quarter the file size of the best direct trace I could get. The catch is that the model moves and resizes the subject a little every run. I measure the box around the ink in both pictures and fit the trace back onto the original.
What I haven't solved is curve fitting. vectorizer.ai gives you real arcs and straight lines. VTracer gives you splines, so circles are never quite circles.
reply