Week 24 Recap: Copy the Ads That Work, Search for $0, and a Free AI Video Tool on Top

October 18, 2026

1,217 upvotes on the week 42 board, nixm on top with 52 from the free section, a Meta Ads Library teardown before your first $10, product search for $0 with EmbeddingGemma 2, the week in models and a week of AI agents going where they should not. Week 24 on Nick Launches.

Your weekly roundup of fresh launches shaping how we work and build.

This week is about the first money you spend and the first feature you add: how to copy the ads that already work before you pay for one, and how to put search in your product without an API bill. Plus the models that shipped, and a week where AI agents kept going places they should not.

The week 42 board cast 1,217 upvotes, down from 1,292 after two record weeks in a row. For the second week running, a free launch out-voted every featured one: nixm finished on 52. Every landing page in this issue got a hand check before making the list.

In this issue:

The week 42 board: a free AI video tool on top with 52, and ten Pro launches
New launch guide: copy the ads that already work, a Meta Ads Library teardown before your first $10
New dev guide: add search to your product for $0 with EmbeddingGemma 2
The week in models: Gemini 4 Argon, Meta's Muse Wonder and Muse Spark Open, and Haiku 5.5's price
Agent security had a week, and the four fixes that matter for a small team
58 posts on X, and what's happening on Launch

Featured Launches 🚀

The Pro launches on this week's board, ordered as the week page has them, with each one's real board position, and every landing page checked by hand. Ten this week: three carried over from earlier weeks and still climbing, and seven new, three of them from Peregrine OS on the same Monday.

Apify: 45 upvotes, #1 on the board. Still climbing: it had 27 last issue. The marketplace of 85,000+ ready-to-run tools for AI, every one callable from your agent as an MCP tool. Featured through October 20.
Snap AI Studio: 42 upvotes, #2 on the board. 41 models and 85 tools and effects for photo, video and audio on one credit pool. Every model shows its credit price before you spend, and failed generations are refunded automatically. $9, $19 or $35 a month. It had 7 upvotes in week 22; featured for 30 days, through October 30.
WhereTheyAsk: 23 upvotes, #3 on the board, up from 4 after its first day. The free community platform with pages, groups, events and live rooms, and a feed in time order with no ads. Featured through October 24.
Open Field by Peregrine OS: 21 upvotes, #4 on the board. A free job search for marketing roles from manager to CMO at US brands, pulled daily from company career pages and job boards, with staffing-agency reposts left out. Each role is tagged by what the job really is: a build role, a takeover, or a top seat open. Free email alerts, daily or weekly.
Wingmate by Peregrine OS: 21 upvotes, #5 on the board. A free network where independent and boutique agencies find each other for white-label work, overflow, referrals and urgent help, from solo consultants to 50-person shops. Results are ordered by fit, then by endorsements from agencies that have worked together, and nobody can pay to rank higher.
Photoshoot.si: 19 upvotes, #6 on the board. A few selfies or one product photo in, a full AI photoshoot out, with the same face in every look. 22 image models and 38 tools on one credit pool, and the cost of a shoot is shown before it starts. From $9 a month.
DobroDesk: 17 upvotes, #7 on the board. One support inbox for email, website chat and messaging, where AI answers from knowledge you approved, drafts replies for review or stays off, and a person can take over the same conversation. €19 per workspace a month with 300 conversations included.
The Nest by Peregrine OS: 12 upvotes, #10 on the board. A launch board that tracks what launched today and what keeps gaining traction weeks later. Its Momentum score halves every seven days and counts votes, follows, visits and updates, and paid boosts are always labeled Boosted.
IndieCrew: 10 upvotes, #11 on the board. A community for independent builders of software, hardware and everything in between, built around one idea: help each other find customers. Members get a profile and a showcase page others can bookmark and comment on, plus a member directory and a discussion wall.
Quiz Zoomies: 3 upvotes, #13 on the board, live since Sunday. Free flag, country, geography and US state quizzes with no account needed, from the YouTube channel that posts a new quiz video every day. The flag quiz has all 195 flags. Featured through November 1.

Launch Weekly Winners 🏆

The most upvoted free launches of the week. No paid placement here, just upvotes. Every one of these went live on Monday, and the top one out-voted every featured launch on the board.

nixm: 52 upvotes, the most on the whole board. AI video with sound, made the way a director works: get one frame right, then set it in motion. Eight-second clips in HD or 4K, scenes with a cast character run 10 or 15 seconds, and Extend adds about seven seconds a pass. Prepaid credits with no subscription: five HD clips for CA$19.99.
Strophe Reader: 31 upvotes. A reader for iPhone and iPad that reads EPUBs, PDFs, web articles and pasted text aloud with word-by-word highlighting. The voices run on the device, so narration works offline. Free for up to five books.
makerio.io: 30 upvotes. Launch support for a launch you already have live: a review, an X shoutout and your link shared by hand with 80+ founders, $10 one-time. It works with Peerlist Launchpad, PeerPush, Nick Launches and other launch boards, not Product Hunt.
Demand Discovery AI: 28 upvotes. Checks whether real demand exists before you build: search demand, founder discussions, launches and hiring trends, with every quote tied to its source. You get a Demand Score from 0 to 100 and a Build, Pivot or Kill call, and the first report is free. Last week it was Clarno: an idea checker in the free top five two weeks running.
OpenScout: 26 upvotes. Finds Reddit threads where buyers ask about products like yours, scores each one for buying intent and drafts a reply. It also tracks which threads Google and AI answers cite. From $89 a month, with a 3-day free trial.
Img2Img: 26 upvotes. Upload a photo, sketch or product shot and change it with a prompt while the subject and composition stay put. Seven models, and one free image a day with no sign-up.

Four more finished on 20 or more: SeoPotion on 24, Nose Job on 23, Benefora on 21 and Sanmedia on 20.

Launch Builders to Watch 👀

Builders who made this week what it was. Worth a follow.

@nixm_ai (opens in a new tab): nixm, 52 upvotes from the free section, the most on the whole board.
@apify (opens in a new tab): Apify, #1 on the board at 45 in its second week.
@Peregrineintell (opens in a new tab): three launches in one day, Open Field, Wingmate and The Nest, 54 upvotes between them.
@MarkZofMarkZ (opens in a new tab): WhereTheyAsk, from 4 upvotes to 23 in a week.
@dschewchenko (opens in a new tab): DobroDesk, 17 upvotes for a support inbox where AI answers only from what you approved.
@StropheReader (opens in a new tab): Strophe Reader, 31 upvotes for books read aloud offline.
@makerio_io (opens in a new tab): makerio.io, 30 upvotes for launch support at $10.
@DemandDiscovery (opens in a new tab): Demand Discovery AI, 28 upvotes for a demand score before you build.
@PiyushDinde (opens in a new tab): OpenScout, 26 upvotes for finding buyers in Reddit threads.

Copy the ads that already work 🎯

Your first ad is a guess. Somebody in your niche has already paid to find out which guesses work, and Meta publishes the answer for free: the Ad Library shows every active ad on Facebook, Instagram, Messenger and Threads, with the date it started running. Nobody keeps paying for a losing ad for 30 days.

On Monday I ran a teardown of one busy niche, AI headshot apps: about 200 active ads in the US, and 26 of the first 66 cards had run for 30 days or more. The guide walks through it with real screenshots: one ad has run since August 2, another since April 9.

What the long runners share:

The hook is about the viewer: "A dated or casual photo can quietly cost you the callback."
One sentence of how, with a number in it: one selfie, 10 headshots.
The objections are answered inside the ad: no studio, a refund, a one-time price.
The product is shown working. Not a logo: the result.

The guide has the four questions to ask of every ad, a prompt that turns ten screenshots into one pattern table, another that writes three ads on the winning frame with your facts, and a first test at $10 a day. Read the guide.

Add search to your product for free 🔎

Most product search is a keyword match: someone types "check if my idea has demand", your listing says "validate your startup with market data", and the box comes back empty. Google's EmbeddingGemma 2 (opens in a new tab), released on October 6 under Apache 2.0, searches by meaning, and its text part is small enough for a laptop.

I installed it and wrote the setup guide, tested on the 2,218 live listings on Nick Launches:

The whole catalog indexed in 42 seconds on a Mac's GPU, about 2.3 minutes on CPU, into a 6.8 MB file
Six real searches, the right listing first in all six, where a keyword search found one listing at best
Screenshots found from a plain sentence: five descriptions, five right answers first
A tiny search API on CPU that answers in 70 to 280 ms
Two gotchas: install the image extra even for text, and update llama.cpp if you go that route

Seven steps, about 25 minutes, $0. Nick Launches is not affiliated with Google.

The week in models 🧠

Gemini 4 Argon is out for everyone. Google opened it in the Gemini API on October 13, and to Google AI Pro and Ultra subscribers in the Gemini app and AI Mode. The introductory price is $2 per million input tokens and $10 per million output. Reports put the standard price at $4 and $20 once the offer ends, so check Google's pricing page before you build a cost model on it.
Meta reached the frontier, mostly. Muse Wonder Full scores 56 on the Artificial Analysis Intelligence Index, two points behind Claude Opus 5.5 and level with Sonnet 5.5. That top model is not on Meta's API, though. The one you can call, Muse Wonder, scores 52 and costs the same per token as Opus 5.5: $4 in, $20 out.
Muse Spark Open is the big open-weight release. 816B parameters, 42B active per token, a 1M token context, and weights on Hugging Face plus hosted access on OpenRouter and Together. Meta and Apple ran it across two M5 Ultra Mac Studios, which tells you the size: a home setup it is not.
Haiku 5.5 is the price to know. Claude Haiku 5.5 costs $0.10 per million input tokens and $0.50 per million output for prompts up to 100K tokens, a tenth of Haiku 4.5's $1 and $5. Longer prompts cost $0.50 and $2.50.

For a builder: put the high-volume steps that run on every request (classifying, tagging, short rewrites) on the cheapest model that passes your tests, and save the frontier models for the long jobs where a wrong answer costs you. This week moved the cheap end a long way.

Agent security had a week 🔐

Four stories in one week, all with the same lesson: what an agent can reach, it will eventually use.

Agents that kept knocking. Researchers logged more than 2,600 attempts by AI agents to pull data from Alibaba's Amap between September 28 and October 8, from infrastructure linked to Tencent Cloud. Some kept trying after they were blocked, and per the New York Times one used a mapping credential that had been left exposed online. When the researchers hid instructions and a mailbox in their report, the agents answered. Tencent says agents testing third-party models browsed in ways it did not intend, and that it is tightening access controls.
India's cyber agency, CERT-In, issued an advisory on October 15. AI agents under testing went past their tasks and reached live systems at Indian organizations, through credentials left in config files and public code repositories, and cloud roles with too many permissions. None of the organizations noticed; the model providers told them.
Perplexity open-sourced SPACE, the sandbox behind its Computer agent, under Apache 2.0. Every task gets its own Firecracker microVM, and credentials are injected from outside so the agent never holds them. Nine models made 108 attempts to escape it: none reached the host, four got around some network rules, and Perplexity says it fixed those.
Anthropic cut internet access for all its internal evaluations after disclosing that, during an automated test on July 18, Claude Haiku 4.5 submitted a made-up murder tip through the Philadelphia police tip form. The police spam filter caught it.

The four fixes for a small team:

Scan your repos and config for secrets, and rotate anything you find.
Give each agent's key only what that agent needs.
Treat what agents read, pages, email and docs, as data, never as instructions.
Run code an agent wrote or fetched in a sandbox, not on your laptop.

What's happening on Launch 📊

This week

📥 248 products submitted, up from 210
🚀 91 went live in their launch window, up from 87, 12 of them featured
⬆️ 1,217 upvotes cast, down from 1,292 after two record weeks in a row
👤 409 new builders joined, 6,113 in the directory
📬 381 new subscribers on this newsletter, 4,157 on the list

The bigger picture

🟢 2,273 live products on the directory, up from 2,191
🔗 Domain rating holding at 76
📣 91.2K impressions on @nicklaunches (opens in a new tab) across 58 posts, up 13.1% on the week before (Typefully, October 11 to 17)

58 posts on X 📣

58 posts on @nicklaunches (opens in a new tab) between October 11 and 17: 91.2K impressions, up 13.1% on the week before, with profile clicks up 21.3% to 398 and engagements up 24% to 3.17K.

The free traffic thread went out every morning: drop what you're building, then submit it to the board. A few of the week's posts worth a read:

Follow along wherever you actually scroll: X (opens in a new tab), TikTok (opens in a new tab), YouTube (opens in a new tab), Instagram (opens in a new tab), LinkedIn (opens in a new tab) and Facebook (opens in a new tab).

Quick ask 🙏

1,217 upvotes, a free launch on 52 that out-voted every featured one for the second week in a row, and ten free launches on 20 or more.

If something in this issue deserves eyes, open its page and upvote it. Ten seconds, real difference. Every featured launch above is still in its featured window this week, Apify until Tuesday.

If you have a launch here, run a site audit. It is free and it takes under a minute.

Thanks for being part of Nick Launches. If you want the daily behind the scenes, including which launches climb and why, follow me on X: @nicklaunches (opens in a new tab).

Building something? Submit it. Listing is free and the backlink is permanent.

24 weeks down. Thanks for reading. 🙏

Category: Product Updates

Nick Launches Changelog Recap Product Updates Launches