Meshy 7 review: how can humans keep up?

The competition among AI 3D generation engines has moved fast enough that keeping a review current is a challenge in itself. Since we started testing Meshy 7, the company has kept shipping new capabilities, and by the time one feature was fully tested, another had already landed. Meshy is scaling at the pace AI companies move at now, a pace few industries, 3D printing included, are built to keep up with. 3D printing has struggled to scale since it was invented nearly 40 years ago; with millions of people now owning a 3D printer and AI increasingly capable of turning a photo into a printable model, that may finally be changing.

Meshy is focused squarely on 3D printing and product design, and the newest Meshy 7 features we tested reflect that focus directly. Multicolor 2.0 exports a textured model as a printable multicolor file. Auto Split separates a model into parts that print and assemble on their own. We also looked at what Meshy calls Alignment, the metric the company uses to describe how closely a generated model matches the image it came from, and at the industry-leading Ultra resolution capabilities Meshy shipped partway through our testing with Meshy 7.1.

Broken geometry, scrambled topology and missing limbs have become rare across leading image-to-3D tools, so avoiding those failures no longer separates one model from another. Failures today are represented by a body slightly too wide, a hand shifted from where the source image placed it, and an engraved line that never made it into the mesh. Nothing looks obviously broken, but the result still isn’t what the user designed. Meshy calls this Alignment and treats it as the metric that decides whether an image-to-3D model is useful for real work rather than merely usable.

Meshy 7.1, released while we were mid-review, is a narrower update aimed at how much fine structure a model’s geometry can actually hold. Its new Ultra 4K mode scales geometry resolution from 2048³ to 4096³ voxels and produces raw meshes of up to 80 million triangles before simplification, a level Meshy says no other 3D generative model on the market has reached yet. To measure the gain, the company built a Detail Richness benchmark that renders a model from multiple angles and tracks how much of its surface needs anti-aliasing at a given screen resolution, a proxy for genuine surface detail rather than agreement with a source image. Against four rival models, Tripo 3.1, Hunyuan3D 3.1, Rodin 2.5 and Hi3D 3.0, Meshy reports Meshy 7.1 ahead at every resolution tested: 53.5% at 1024 (2.1 points over Rodin 2.5), 37.4% at 2048 (2.4 points over Rodin 2.5) and 23.1% at 4096 (1.2 points over Hi3D 3.0). Ultra 4K is currently limited to single-view generation and to paid tiers; Ultra 2K, also new, works with both single and multiple views.

Pricing hasn’t moved: a standard generation still costs 20 credits, and output exports to eight formats, from GLB and FBX to STL and 3MF, with one-click plugins for Blender, Unity, Unreal Engine, 3ds Max, Maya and Godot. Models can also be sent straight to the most popular slicers, including those from Creality, Bambu Lab, Snapmaker, Elegoo and Ultimaker.

Those are Meshy’s own numbers and claims. What follows is a hands-on account of whether Meshy 7 and Meshy 7.1 hold up under scrutiny and printing in our own lab, tested against the kinds of assets the company points to as its strongest cases: organic characters, car models built and printed from multiple images and in multiple parts, and a handful of display and functional models where accuracy and alignment decide whether the parts are actually usable.

Speeding into 3D generation

Hypercars seemed the obvious choice when we went looking for objects that would test both accuracy and extreme geometry at once. The project started with a Lamborghini parked in front of a garage near our offices. We photographed it from four angles: main view, left side, back and right side, to test Meshy’s multi-view capability. The result was excellent everywhere except the exhaust, which is black on black at the rear: the first generation didn’t distinguish the two pipes but it did catch many of the color details, including the red Brembo brakes and rear lights, but a second attempt reproduced both extremely accurately.

We then ran the model through Multicolor. The result was striking and reasonably accurate: not a precise replica, but close enough to pass for a concept sketch, the kind of rendering a design team might use at the start of a project. We printed the car both in multicolor filament and in high-accuracy stereolithography resin. Read on to learn more about the printing results further down.

For another test on a European concept car, we modeled the Cupra Dark Rebel concept, this time from a photo taken a few years ago at an event during Milan Design Week; we picked Cupra specifically because the company has an active interest in 3D printing for interior parts. Once we had the model from a single photo, we moved on to testing Meshy’s printing features directly. We recreated the model in color and also split it up for printing separate parts. The system automatically separated the wheels from the body for easier printability.

Next up, from photos taken of a BYD U9 on display at Tel Aviv airport, we reproduced the car, again from multiple angles and at 2K resolution. We also tried Ultra 4K from a single image this time: every dimension came out accurate, down to the detailed geometry of the rims. We ran this model through Auto Split too. The automatic setting divided it into four parts: wheels, upper body, lower body, and roof, and laid them out on a plate ready to print through any of Meshy’s integrated slicers, from Bambu Studio to Ultimaker Cura, or ready to bring into Blender or Unreal Engine for further work. We also generated a textured, multicolor version, covered later in the printing section.

Working with another Chinese-made hypercar, we reproduced a Xiaomi Vision GT concept, one of the fastest hypercars ever built, this time from a single image found online. This model came out even more accurate than the previous ones, likely because its softer, simpler lines suit the AI better. We generated it both in high resolution and as a textured model for color printing.

One of the strongest results came from the Czinger C21, a car largely built for digital manufacturing since most of its frame and internal components are 3D printed and its exterior is composite. The multicolor version reproduced the car’s silver finish well, though some of the more complex details, one of the rims and especially the latticed rear grille, were too intricate for the system to resolve fully. Overall, however, the concept looked great with the car’s unique shape and dimensions recreated faithfully.

Meshy offers several ways to get from model to print. The system can repair a model for printability first, then export it as a single-color STL for most stereolithography slicers or as a 3MF file for most filament slicing systems. Users can also send it directly to one of the popular slicers in a couple of clicks.

Multicolor printing runs through its own option, which opens a choice between a more realistic rendering and a cartoon-style one; users select the Print in color option, choose color count and noise reduction levels, and send the file directly to their favorite slicer.

New and more accessible multi-color material jetting systems are on the way, from Flashforge and Sailner, but today these systems are still mostly for professional studios, along with Mimaki and Stratasys polyjet machines. Meshy formed a special partnership with Flashforge, with the company integrating Meshy into its Flash Studio software. Flashforge also said it is shipping its $3,999 CJ270 multicolor material jetting 3D printer on October 15th. This could be a major breakthrough for Meshy’s multicolor capabilities, as the Flashforge system, like other material jetting systems, can print in color directly from any OBJ file export.

For now, unless a printer has material-jetting multicolor capability or one of the newer CMYK systems from Snapmaker or Bambu Lab, the cartoon setting is more than sufficient. From there, the number of colors can be set, typically four, matching most standard multicolor filament systems on the market. The finished file goes to the slicer, where colors can still be adjusted before printing.

The results were solid all around. Printing an original multicolor model is no easy task, given how many filament changes it demands. We used a Bambu Lab H2C for the first two prints and ran into some issues, mostly with supports detaching rather than problems with the model itself. Both the first prints, the Lamborghini and the Xiaomi, came out successfully on the second try. The colors were accurate for a filament system, which says as much about the accuracy of the source model as it does about the printer; the result isn’t comparable to a die-cast or an industrially finished model, but on dimensional accuracy and detail there was little to complain about. The high-resolution stereolithography prints looked even better. We did that for the Lamborghini second version and for the ultra-high-resolution BYD U9.

Putting physical AI to work

The next test moved into functional and decorative parts. The first was a replica of a serpent statue, the mascot of Inter Milan, which we photographed at the club’s store at the stadium, two photos in total. Turning those into a clean 3D replica took Meshy seconds. We printed it both as a single piece, which came out well, and through Auto Split, which produced three parts: the serpent, the Inter Milan logo and the base. Splitting it wasn’t strictly necessary, since it barely reduced the supports needed, but it would make sense for anyone planning to paint the three parts in different colors. Assembly was simple, with glue needed only to attach the serpent’s body to the base.

For another test we had to replace two broken hooks from a wall rack that originally held four; two of the wooden hooks had snapped off. We photographed the hook and took a few measurements, and Meshy reproduced the part accurately. Setting the correct height in the slicer was the only adjustment needed before printing. The part wasn’t dimensionally exact to a rigorous engineering standard, but every dimension we checked came out accurate to the millimeter: setting the height at 9 cm gave a top diameter of 4.3 cm, matching the original hook exactly. The model even reproduced a small cut at the top of the original, adding a layer of realism. It still needed paint to match, but the base result held up well.

We also modeled a Nike shoe for a separate ongoing project, working from a handful of images downloaded from different angles. The file came out accurate except for one small detail on the shoe’s upper section; the laces, the Nike logo and even the outsole were reproduced in fine detail. To test printability and resolution, we printed it on both a Form 4, using Formlabs’ Precision Model resin, and on the Elegoo Jupiter 2, which we’re reviewing separately. The Form 4 result returned, as expected, excellent results in terms of accuracy; the Jupiter 2 print was our first on that printer, and the result was also impressive, although some layer lines were more visible on a much larger model.

The last set of tests involved robots and action figures. Months earlier, an older version of Meshy had already produced accurate replicas of Japanese anime robots from the 1980s using images found online. This time, we used Meshy 7 to replicate the giant Gundam statue in Tokyo from a photo taken in summer 2025, a shot that had given earlier versions of Meshy and other AI 3D tools trouble with perspective. The new model came out accurate, aside from one extra detail on the hand that needed removing in post. We printed it two ways: as a single piece in resin, and through Auto Split in filament, which split the model at the leg joints and laid the parts out ready to print (will add more detail on results as soon as prints are completed).

Discovering a business opportunity

Meshy’s resolution and accuracy gains are real, and so is the pace at which the system keeps improving. What changes fastest are the small things: exact lettering, logos, precise color shades, and how well the AI fills in the parts of an object hidden from the camera. To close out testing, we ran two lighter projects: turning one of the robot models into a chibi figure and printing it in multicolor and creating chibi figures from photos of a few different people.

The range of what Meshy can produce is wide, and its ease of use is amazing, but it’s worth being precise about what it is. Meshy 7 produces visually faithful 3D, not CAD: it’s built for how an object looks, for visualization, concepting, game assets, 3D printing and decorative engraving, not for dimensionally accurate, tolerance-correct engineering parts or functional prototyping. Humans can keep up by inputting new ideas and adding engineering capabilities, working with Meshy to expand their capabilities and horizons.

Within that scope, it works both as a starting point for a more demanding engineering project and as a capable tool for decorative objects, toys and simple functional parts, with Meshy 7.1’s resolution gains and Auto Split’s part-splitting already changing what counts as printable straight out of the box.