Three ex-Ramp engineers just raised $25M for Melius, an AI ad creative platform that turns a text prompt into finished ad campaigns — images, video, and copy — after scrapping their first product entirely. The round ($20M Series A led by CRV, on top of a $5M seed from General Catalyst) landed October 6, 2026, and Melius says it crossed $1M in annualized revenue within two months of its July launch. If you're paying a creative team or freelancer by the asset for Meta or TikTok ad variants, this is the clearest signal yet that the cost of "just make me ten more versions" is about to fall.
What actually happened#
Joowon Kim, Young Kim, and Arnav Ramu left Ramp, the corporate spend-management company, to build an AI tool for managing ad spend. Six months in, they killed it. "We scrapped the entire codebase; we burned all of it," Kim told TechCrunch. They rebuilt around a different bet: an "agents lab for creative work" where you brief an AI agent in plain language and it assembles images, video, and campaign copy for you to approve or steer. It launched in July 2026 and apparently found product-market fit fast — $1M ARR in two months is a genuinely unusual ramp for a brand-new creative tool, and it's already competing against Higgsfield (reportedly $700M+ ARR), Krea, and Flora AI in the same space.
Does AI ad creative actually change anything if you're not running six-figure budgets?#
If you're spending under a few thousand dollars a month on Meta or Google ads, no — this changes nothing for you yet. Pricing and access details for smaller accounts aren't public, and the tools competing here (Higgsfield, Krea, Melius) are chasing agencies and bigger D2C budgets first.
Where it does change something: if your biggest line item in ad spend isn't the media buy, it's the creative — a designer or video editor billing you per variant so you can test five hooks instead of one. That's the cost this category is squeezing. We've watched this exact shift happen with still-image generation over the past year (GPT-Image-2.5's launch in September was the same story for product photography, and ChatGPT's visual ads push is the distribution side of the same trend), and video creative is now going through the same curve, just twelve months behind.
The honest trade-offs#
The upside is real: faster creative testing cycles, lower cost per variant, and — if Melius's ARR number holds up under its next funding round's diligence — a tool category with actual paying customers, not just a demo. A founder who burns a working product because the market signal was wrong, and rebuilds in under a year to $1M ARR, is a team that can execute. Rare, in this market.
The honest unknown is quality consistency at scale. Every AI creative tool we've tested produces great hero shots and inconsistent everything-else: brand colors drift, product proportions shift between variants, and on-model human photography still needs a human pass before it ships. None of the public reporting on Melius addresses retention or output quality at volume, only the revenue ramp. I'd treat the $1M ARR figure as "people are willing to try it," not "people are happy a year later" — we don't have that data yet, and neither does anyone citing this story this week.
How we're building around this#
At WebEpex we run Meta ads creative production through a pipeline that mixes GPT-Image-2.5's Flare model for product stills with manual video editing, because we've found the gap between AI-generated and human-finished video is still wide enough to matter for client-facing ads. We reviewed Melius and two other node-based creative-agent tools the day the funding story broke.
We're now running a small internal test batch against our own current pipeline before any client campaign touches it — same hook, same audience, measuring output quality and iteration speed against what a human editor produces in the same time. That's the bar: if an agent tool can't beat a skilled editor on speed without losing quality, it doesn't replace the editor. It just becomes another step in the review queue.
We build for GCC, European, US, and Canadian ad accounts mostly, with Indian SMB clients increasingly asking for the same creative-testing speed now that the tooling exists. It's also worth keeping an eye on the permissions side of this if you're letting any AI tool touch a live campaign — Meta's own AI connector permissions default more openly than most advertisers realize. The point worth making plainly: we're not against AI creative tools, we're against swapping in a cheaper tool and quietly accepting worse output — those are two different decisions, and this round of funding doesn't change which one we'd make.
What I'd tell a client asking about this this week#
Don't switch your creative pipeline over a funding announcement. Do ask whoever's producing your ad creative right now — in-house or agency — whether they're already testing tools like this, because if they're not, they're about to get undercut on turnaround time by someone who is. If you're running under 10 ad variants a month, this is a "watch, don't act" story. If you're past 50 variants a month and still paying per-asset, it's worth a real cost comparison within the next quarter, not today.
If you want a second opinion on whether your current ad creative spend is actually the bottleneck, send me what you're running and I'll tell you straight — takes two minutes and you don't have to buy anything. cal.com/webepex/growth-review
