1.3 stars, 150k installs: an AI car-mod app and the Vietnamese paid-acquisition machine behind it

"Midas: AI Car Modification" is an AI car-mod preview app. Take a photo of your car and it generates a modified render. Less than two months after launch, it reached 157k installs on Google Play with a 1.3-star rating. It got there not through the product but through an old playbook from a Vietnamese studio: multiple developer accounts, batched paid acquisition switched on and off, and watch-ads-for-generations. The company has now launched two more identical packages to take over. Below: how it got going, where the money comes from, and whether this path is worth taking.

A car that was swapped out

On Oct 13, a user named Chris Dry left a review on Google Play: "I uploaded my car, picked the mod items and color, and it generated a completely different car, with a different color. Nothing to do with what I asked for." 10 people hit "helpful." A week later, on Oct 6, another user was blunter: "Too many ads. After watching the ad, it swapped my car for another one. Garbage, the free AI is much better." On Oct 12 there was another: "I watched six or seven long ads, and it generated a random picture from the web and passed it off as my modified car." (Google Play reviews, read 11-06)

By Nov 6 it had 344 ratings on Play, 303 of them one-star. That same day its cumulative installs were 157,320, up 43,776 from 7 days earlier.

What it is

The feature does one thing: upload a car photo, pick from 10 mod styles (paint and wraps, wheels, body kits, spoilers, lowered suspension, underglow), and the AI renders it. Category is Art & Design. It launched in Sep 2026 10, developer credited as Echola AI. The store page says "Contains ads" and "In-app purchases," with single items at $9.99–49.99. According to the reviews, free users have to watch several ads before each generation.

The store listing and package name are stuffed with keywords: the package is com.midas.ai.modifycars.carmodification.car, and the description repeats car modification, tuning, body kit, rims, aiming at searches like "ai car modification".

Where the volume comes from: the curve gives it away. It's bought.

Organic growth doesn't look like this. Using Google Play's public install counts and AppstoreSpy's estimated daily installs: in Oct 3 and 4 days, it added 2,575 and 4,285 per day, then dropped to 6 on 5; on Oct 13 it added 12,118 in a day, 14 on 7,289, and 18 on 15; in October 27 and 28 days it was 8 a day, then by 29 to 31 it was back to about 7,150 a day. Thousands one day and single digits the next looks like the pattern of ad budgets switched on and off in batches (inferred).

Other details fit too:

• Every country it charted in was a cheap-acquisition market. In Nov 6 it ranked #40 in the Thailand Art & Design Top Free chart, #87 in Mexico, #93 in Vietnam, #99 in the UAE, and #100–200 in Argentina, Australia, the Philippines and Saudi Arabia, with no Top Grossing rankings at all.
• The developer's site midasstudio.app is hiring for a User Acquisition role, and the fourth step in its own described workflow is "ASO & growth".
• There's a dedicated Facebook page for "Midas: AI Car Modification", a setup common for small apps running Meta ads. Whether the spend went to Google or Meta, and how much, isn't visible from outside (unverified).
• Content channels are nearly absent. There are plenty of TikTok videos on "AI car mod", but the ones mentioned are Modsnap and "AI Car Designer - Tuner" on iOS (the latter posted a creator video with a promo code in Nov 1). I found no one mentioning Midas. Nothing on X, Reddit or in the press.

One more small oddity: on Oct 30 it gained 17 five-star ratings in a single day, and since then its one-star and five-star counts have stayed identical (20 each). Whether these were fake, there's no direct evidence.

Who's behind it: one email, three accounts, ten apps

10 apps on Play list the support email [email protected], spread across three developer accounts: Midas Studio App, Echola AI, and MEDICI TECHNOLOGY GROUP. The developer entity on the store page is MEDICI CORPORATION JOINT STOCK COMPANY, based in Ho Chi Minh City. The careers page lists a hybrid role in Hanoi; the rest are remote.

(Google Play public data, 11-06)

Wallpapers, lock screens, cleaners, QR scanners, AR tracing, plus AI car modding: all old-school ad-monetized utility app categories, with a new app every week or two. The most telling are the two bolded rows in the table: Car Modify (Oct 26) and CarDes (Oct 27, package still com.midas.ai.cardes...) do the same thing as Midas. On Oct 29 they had 46 and 4 installs; a week later, 55k and 47k, still with no ratings, and they had already charted in 19 and 24 countries (CarDes hit #27 in South Africa's Art & Design), more than the 8 countries Midas itself charted in. The likeliest explanation (inferred): 1.3-star ratings killed Midas's store conversion, so the budget moved to clean new packages. Midas's drop to just a thousand or two new installs a day in November 4 and 5 fits this too.

Where the money comes from

Mostly ads. Free users watch several rewarded videos per image generated, and in markets like Thailand, Vietnam, Mexico and the Philippines install costs are low. As long as ad revenue per user exceeds what it costs to acquire them, it can keep buying installs (inferred). The $9.99–49.99 in-app purchases are a side line; it has no Top Grossing ranking, so IAP revenue is likely small.

This playbook never depended on retention: when one package's rating collapses, swap in another and keep buying. The company's sister wallpaper and lock-screen apps sit above 4.4 stars with hundreds of thousands to millions of installs, showing the pipeline runs well in categories that don't need "AI results". AI car modding is the first time it ran into a category where bad generation quality gets punished in reviews.

Who it's competing with

AI car modding was already crowded in 2025–2026. Midas is clearly chasing others:

(Android data from Google Play public data; iOS figures are App Store ratings counts, 11-06)

Functionally Midas has no edge over them, and its generation quality is actually the worst. Its only real difference is the business model: the others charge car enthusiasts through subscriptions or IAP, while it charges everyone who clicks in through ads.

If it were you

Verdict (C): The app itself isn't worth following. What's worth learning is the machine behind it. Demand for AI car modding is real (ModsTune has 620k installs and 4.55 stars after three-plus months), but a dozen-plus players are already in it, the top one has good quality and no ads, and general AI can do the same thing for free. Midas got traction thanks to a studio's ready-made acquisition team and account network, which someone following later won't have.

Who it's for: Small teams with paid acquisition and ad monetization experience and a pipeline of "buy one package, break even, swap to the next", treating AI car modding as just another category to test. Or people who can get generation quality to the point of "it still looks like my car" and build the top-quality tier of the category.

Who it isn't for: Indie devs who've never run ads and are counting on organic store traffic (Midas's organic is close to zero, with search slots taken by a dozen players). Also anyone planning a thin AI wrapper that trades ads for generations (the top complaint is "free AI does it better", and a 1.3-star package won't survive two months; Midas itself has already been replaced).

How to do it: If you must enter, don't copy Midas, copy ModsTune and fill one gap they all leave open. First, the car must not change: use an image editing model to alter only the specified parts, keeping the model, angle and plate intact. This is the most common complaint in reviews. Second, make the results actionable: after generating, provide rim sizes and wrap color codes, and connect to local body shops or parts suppliers for quotes. That gives a chance to move from "playing around" to "people who really plan to modify their car are using it", and a chance to charge the shops (inferred).

How to acquire users: Two paths. If you can run ads, follow Midas's approach with small budgets in cheap markets like Thailand, Mexico, the Philippines and the Middle East, and first work out how many ad views a user must generate to break even. If you can't, go through car enthusiast communities on TikTok: partner with car influencers who post before-and-after videos of their own cars. On iOS, products already use creator videos with promo codes, so the path is proven to work.

When to get in before it closes: There's no hard deadline, but the window is narrowing. Each update to ChatGPT's or Gemini's image editing cuts into the value of single-function apps. Midas alone shipped 6 copycat packages in 3 weeks, meaning cheap-market search and ad slots will keep getting pricier.

What would prove this read wrong: (1) If CarDes and Car Modify drop below 2 stars a month later and daily installs fall to a few hundred, even a team with a working acquisition machine can't break even, so the ad model doesn't work in this category. (2) The reverse: if they still add tens of thousands a week in December, ad break-even holds, this company keeps adding packages, and what you can copy is the playbook, not the category. (3) If ModsTune's weekly installs are cut in half within 30 days, demand is being absorbed by general AI and the whole category is in retreat.

Data and sources

• Installs, rating counts, star distribution and chart rankings: Google Play public data, 2026-10-29 to 11-06; daily installs since launch are AppstoreSpy estimates.
• Store listings and reviews: play.google.com/store/apps/details?id=com.midas.ai.modifycars.carmodification.car (read 11-06, including developer entity and address); store pages of the company's other apps.
• Developers: midasstudio.app and ecola.ai (read 11-06); Facebook page "Midas: AI Car Modification".
• Competitors: Google Play and App Store public data (11-06); TikTok search for "ai car mod" (11-06).
• Where you see "inferred", "estimated" or "unverified": these are inferences, estimates, or places where I found no direct evidence. Acquisition channels, spend and break-even figures have no public data.