Withdrawal and Cancellation of the Fulbright-Hays Group Projects Abroad Program [pdf] (public-inspection.federalregister.gov)

Hey HN! We’ve been experimenting with integrating multimodal models directly into creative workflows, and ended up building an AI-first image editor using OpenAI’s new `gpt-image-1` (from GPT-4o) inside our SDK.

Instead of prompting in ChatGPT and pasting outputs into a design tool, this lets you generate, edit, and remix images all in one canvas. This allows for really interesting new workflows, like quickly mixing multiple images, or creating visual prompts by using annotations and reference on the canvas.

Some key details:

- Built with our plugin system in CE.SDK (CreativeEditor SDK) - Works with any multi modal model/API, not just OpenAI - You can bring your own API key, or extend it however you like - Focused on real-world workflows, not just concept generation

We put together a live demo and wrote a bit more on why we built it.

Would love feedback or to hear how others are thinking about AI in creative tools. If you have questions or ideas, let us know!

Comments (7)

olcioszek · 4h ago

I'm curious about the architecture behind this integration — how does the system handle communication between GPT-4o and the visual editor in real time? Are there any challenges in syncing AI-generated content with editable UI elements?

OriginalMrPink · 4h ago

impressive...can't wait for OpenAI to make their image gen faster tough

klaudial · 4h ago

Cool! How does this compare to what tools like Canva or Adobe offer for automation?

buss_jan · 3h ago

Well, it's fundamentally inspired by what these companies are trying to do. However, we are an SDK so we want to offer a toolset for other apps to build similar experiences.

GoezdeB · 4h ago

Really interesting! I noticed it takes quite a long to generate images, what’s your take on managing latency?

buss_jan · 3h ago

Hey, yes, that's somewhat unavoidable given that is simply takes OpenAI that long to generate images. I think that's to be expected by most users. We tried to be as friendly as possible to parallelization so users aren't blocked while working in the editor.

nablahero · 4h ago

I would assume openAI bring in the delay?

E-bikes make your city smaller (mattsayar.com)

How to Achieve Structured Output in Claude 3.7 (baz.co)

I'm Worried About Junior Developers (envylabs.com)

PostgreSQL 18 Beta 1 Released (postgresql.org)

QueryLeaf: SQL for Mongo (github.com)

High-Temperature Superconductivity (en.wikipedia.org)

Huawei unveils laptop running self-developed HarmonyOS as Windows licence expire (scmp.com)

Parse_searchable_rolls – Parse Searchable Electoral Rolls (github.com)

Why Noisy Rooms Are Harder for Some Brains to Handle (neurosciencenews.com)

Show HN: I Built a TypeScript Game (atypescriptgame.com)

Withdrawal and Cancellation of the Fulbright-Hays Group Projects Abroad Program [pdf] (public-inspection.federalregister.gov)

Shiny New ZORA Token Loses Its Spark as Traders Remain Wary of VC-Backed Tokens (coindesk.com)

California state bar sues (remote proctoring) vendor after troubled exam (abajournal.com)

Path Tracing a Trillion Triangles at Intel (community.intel.com)

Show HN: I crammed my daily tools into a single iPhone keyboard (flexiboard.pressdeck.io)

MCP Server Support on Vercel (vercel.com)

Remote DoS in a specific Litespeed environment via a single curl request

Once-Weekly Semaglutide in Adults with Alcohol Use Disorder: A RCT (jamanetwork.com)

Bungie vaulting Destiny 2 content backfires with the latest lawsuit court ruling (windowscentral.com)

Show HN: A Coilgun I Built When I Was 16 (github.com)

Bill Gates pledges 99% of his fortune to Gates Foundation (apnews.com)

Management Cybernetics 101 for Urbanists and YIMBYs (governance.fyi)

How AI Demand Is Draining Local Water Supplies (bloomberg.com)

Why Walking Helps Us Think (newyorker.com)

Elon Musk is responsible for "killing the poorest children," says Bill Gates (arstechnica.com)

Watch Humanoid Robot Attacking Workers During Testing in Chinese Factory [video] (youtube.com)

A Politically Neutral Military Is Not Always Obedient (lawfaremedia.org)

20 years to give away virtually all my wealth (gatesnotes.com)

Vibe Coding MenuGen (karpathy.bearblog.dev)

Trump's NIH Axed Research Grants Even After a Judge Blocked the Cuts (propublica.org)

Why Germs Love Our Collective Amnesia (nautil.us)

Datasets Are All You Need (ai.intellectronica.net)

6502 Illegal Opcodes in the Siemens PC 100 Assembly Manual (pagetable.com)

OpenAI hires Instacart CEO Fidji Simo as head of applications, reports to Altman (openai.com)

Salt is the fastest, most intelligent and scalable automation engine (github.com)

Softlanding Linux System (en.wikipedia.org)

Arista cats purr over $2B quarter while tariff time bomb ticks (theregister.com)

Why do so many people hate Elon Musk? (msn.com)

Adventures in Imbalanced Learning and Class Weight (andersource.dev)

Free Access to QNX for Non-Commercial Use (2024) (qnx.com)

Why WebAudio Isn't Enough for Serious Apps (switchboard.audio)

Show HN: DevDocsDown – Instantly Download Dev Docs (devdocsdown.xyz)

Creating Furniture-Scale Objects with a Computer-Controlled Sewing Machines (drive.google.com)

JSON Schema (json-schema.org)

Why AI agents need a protocol like MCP to reach their potential (sdtimes.com)

Google to Back Three New Advanced Nuclear Projects (esgtoday.com)

Why the rich paid less tax in the 1970s – despite 98% tax rates (taxpolicy.org.uk)

Ask HN: Anyone using Figma Make? What have you built?

Appeals Court Sidesteps the Big Questions on Geofence Warrants (eff.org)

TLDR: Official MCP Registry (mcps.sh)

Show HN: An AI-first visual editor using GPT-4o's GPT-image-1 model

Comments (7)