惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

WordPress大学
WordPress大学
腾讯CDC
阮一峰的网络日志
阮一峰的网络日志
GbyAI
GbyAI
B
Blog RSS Feed
Engineering at Meta
Engineering at Meta
Google DeepMind News
Google DeepMind News
MyScale Blog
MyScale Blog
Last Week in AI
Last Week in AI
F
Fortinet All Blogs
云风的 BLOG
云风的 BLOG
N
Netflix TechBlog - Medium
G
Google Developers Blog
博客园_首页
有赞技术团队
有赞技术团队
V
V2EX
奇客Solidot–传递最新科技情报
奇客Solidot–传递最新科技情报
MongoDB | Blog
MongoDB | Blog
H
Help Net Security
aimingoo的专栏
aimingoo的专栏
月光博客
月光博客
Hugging Face - Blog
Hugging Face - Blog
The GitHub Blog
The GitHub Blog
S
SegmentFault 最新的问题

Hacker News - Newest: "AI"

AI can't read an investor deck AI as an attorney? Student uses ChatGPT, Gemini to sue UW over alleged racial discrimination Hacking MCP Servers in AI Systems – The Rug Pull: Tool Changes After Approval GitHub - MeepCastana/KubeezCut: Free Web based video editor Can AI judge journalism? A Thiel-backed startup says yes, even if it risks chilling whistleblowers Coming soon: 10 Things That Matter in AI Right Now DARPA built an AI to fact-check enemy weapons claims What explains heterogeneity in AI adoption? When AI Meets Muscle: Context-Aware Electrical Stimulation Promises a New Way to Guide Human Movements - Department of Computer Science AI Changed How We Build. It Did Not Change What Matters. Linux rules on using AI-generated code - Copilot is OK, but humans must take 'full responsibility for the… Meta spins up AI version of Mark Zuckerberg to engage with employees Code Mode: Let Your AI Write Programs, Not Just Call Tools | TanStack Blog GitHub - Delavalom/graft: Go framework for building AI agents. Type-safe tools, multi-provider (OpenAI, Anthropic, Gemini, Bedrock), zero vendor SDKs. India's TCS tops estimates, says new AI models did not dent services demand Gen Z's fading AI hype Strong feeling: we are in a folded AI reality GitHub - machinarii/total-recall-catalog: A reference catalog of latest knowledge retrieval, memory & RAG systems GitHub - mensfeld/code-on-incus: Give each AI agent its own isolated machine with root, Docker, and systemd. Active defense detects and stops threats automatically.. Quantization, LoRA, and the 8% Problem: Benchmarking Local LLMs for Production AI Iran war: We spoke to the man making Lego-style AI videos that experts say are powerful propaganda Powell, Bessent discussed Anthropic's Mythos AI cyber threat with major U.S. banks GitHub - immartian/bellamem: Persistent belief-graph memory for AI agents. Retrieves decisive context by importance — not recency, not RAG, not /compact. recursive-mode: The Repo-Native Operating System for AI Engineering After the attack on Sam Altman's home, will AI CEO's go on the offensive? The biggest advance in AI since the LLM Opus 4.6 vs GPT 5.4 One Prompt Unity World Generation Test “AI polls” are fake polls Client Challenge Can AI be a 'child of God'? Inside Anthropic's meeting with Christian leaders
Creating SVGs with AI
Gabriel Furstenheim · 2026-06-26 · via Hacker News - Newest: "AI"

SVGs are great in general, light, vectorial... In a world of AI-generated content, they are especially practical: AI content is great, but it always has some artefacts, and it isnt great at maintaining consistency. If we have SVGs, we can fix those tiny defects and ensure consistent assets across our application.

AIs are extremely good at creating images, and somehow fail miserably when creating SVGs. Theres a rumour that Fable is good at it, but who can actually verify that? The good news is you dont need it to create great, simple, and consistent SVG icons for your application. Heres how to do it:

The process to obtain SVGs is dead simple:

  • Create PNGs

  • Vectorize the PNGs

Instead of generating SVGs with AI, we will ask the AI to generate raster images that look like SVGs. Beyond that, there are a couple of subtle tricks at play here:

  • Use Codex for this task; it’s miles ahead of Claude Code.

  • It’s super important that we forbid the AI from even trying to create an SVG.

  • Split the task into two agents, the judge and optimizer, and the executor. The main agent judges the quality and consistency of the images, and optimizes the prompt. This prompt is handed to a subagent that does the image creation. I believe this is in essence what “looping” is about, but with Codex it works out of the box without using /goal or /loop.

The actual prompt for my PNGs was the following (it was a one-shot):

The end goal is to generate some images for me. This will be a family of svgs that I want to have a similar style and motif, so you should do in two phases. You should optimize the prompt that you later handle to a subagent. Iterate until the result is good. You will create a clean svg-like icon style sheet. The goal is generated icons that look like polished vector SVG marks, not hand drawn chunky svg paths and not detailed AI concept art. Each svg will be represented in a consistent shared circular segment ring. Once you have this. The end goal is to have a png for each combination of animal in animals.txt and instrument in instruments.txt, for example an otter playing a violin

There are a lot of tools to convert a PNG to an SVG. But there is only one that works: vectorizer.ai.

As a quick experiment, I ran the same image through three different tools. The difference is abysmal. This comparison is not to bash anybody; vectorizing is a seriously complicated problem. In fact, the owner of vectorizer.ai has a PhD from Stanford on this topic.

As you can see, creating SVGs is just a matter of using the right tools for the job. You can check all the animal and instrument combinations here!

Discussion about this post

Ready for more?