STL Viewer: Meshy AI vs. Snapshapy Normal Map Test

STL Viewer: Meshy AI vs. Snapshapy Normal Map Test

STL Viewer, Meshy 3D Agent, Snapshapy

If you use an stl viewer to inspect downloaded 3D assets, this comparison helps you check whether Meshy 3D Agent and Snapshapy's 3D Agent accurately describe triangle counts and texture maps in their GLB files.

I asked Meshy 3D Agent for a lightweight, game-ready treasure chest for a mobile game, and it replied that the chest was "ready at 4,962 triangles, keeping the full texture fidelity, normal maps, and silhouette intact." The triangle count was true to the digit. The normal map was not there: the mobile GLB I downloaded holds two textures, base color and metal/roughness, both still 4096 × 4096, and no normal map, although the 91.4 MB original had one. Snapshapy's 3D Agent warned me before its job ran that it might stop short of its target, then printed "8.7k triangles · seams stopped it above 5.0k" on the result card. Its file carries a 1024 × 1024 normal map.

This review is about one question: does each agent's reply describe the file it hands you? I tested Meshy 3D Agent at meshy.ai and Snapshapy's 3D Agent at snapshapy.com on my own paid accounts, Meshy Pro ($20 a month, 1,000 credits) and Snapshapy Pro. I checked every reply against a balance, a job card, a stopwatch, or a file I downloaded and opened myself.

Hitting a number and reporting a number honestly are different skills. In this test, the two agents split them.

My verdict in 60 seconds — Meshy 3D Agent 60/100, Snapshapy 89/100

Meshy 3D Agent delivered the triangle count it promised, but its reply listed a normal map the file does not contain, while Snapshapy missed its triangle target and said so on the card before I could find out myself.

  • Reply matches the file — Meshy 4/10 · Snapshapy 9/10 — Meshy named normal maps the GLB lacks; Snapshapy's card printed 8.7k against its own 5.0k
  • Triangle count vs promise — Meshy 9/10 · Snapshapy 5/10 — 4,962 exactly as stated vs 8,654, 1.73× its own 5,000 target
  • Normal map in the mobile GLB — Meshy 1/10 · Snapshapy 9/10 — none vs 1024 × 1024, though Meshy's 91.4 MB original had one
  • Mobile texture budget — Meshy 3/10 · Snapshapy 9/10 — 33,554,432 texture pixels vs 3,145,728, 10.7×
  • Warning before the job — Meshy 3/10 · Snapshapy 9/10 — no caveat vs "can stop short of a low target on seam-heavy models"
  • Stated cost vs charged cost — Meshy 9/10 · Snapshapy 9/10 — Meshy's Est. 36 and Est. 30 matched its balance; Snapshapy's 8 and 3 matched "11 spent"
  • Approval stops — Meshy 5/10 · Snapshapy 9/10 — 2 Approve clicks vs 0 across the same 3 messages
  • Credits per mobile chest — Meshy 4/10 · Snapshapy 9/10 — 66 vs 11, $1.32 vs $0.52, 6.0×

The concrete problems with Meshy:

  • The message 3 reply says normal maps were kept; the downloaded mobile GLB has none.
  • Both Meshy textures stayed 4096 × 4096, so 36.6 MB of the 37.0 MB file is texture, 99 %.
  • The reply gave no warning about either point, so I only learned them by opening the file.
  • Meshy charged 66 credits and stopped twice for approval on the way.

For a mobile game prop, I would use Snapshapy.

Same three messages, typed word for word

Each agent got one conversation and the same three messages:

  1. "Give me 4 concepts for a fantasy treasure chest game prop."
  2. "Go with option 1 and make it a textured 3D model."
  3. "I need a lightweight game-ready version for a mobile game."

Whenever an agent showed a choice card, I picked the first option. Meshy started at 856 credits, Snapshapy at 1,046. I ran the Snapshapy half the next day, after an update; the Meshy numbers are from the first day.

Meshy 3D Agent's chat home before the test.

Snapshapy 3D Agent home with 1,046 credits before the test.

Message 3 is the one that matters here. "Lightweight" and "game-ready" are both claims an agent has to make about a file, and a mobile build lives or dies by what is actually inside that file.

Meshy's message 3 reply, claim by claim

Meshy answered message 3 with no approval card and no charge (790 before, 790 after). About 117 seconds after I sent it, this arrived: "The mobile-optimized version of the Dwarven Runic Chest is ready at 4,962 triangles, keeping the full texture fidelity, normal maps, and silhouette intact for real-time mobile rendering."

Meshy message 3 reply: "ready at 4,962 triangles, keeping … normal maps … intact".

That sentence makes four claims, and I checked each one against the Meshy GLB:

  • "4,962 triangles" — Meshy file 4,962 triangles, 8,566 vertices · matched exactly
  • "full texture fidelity" — Meshy file 2 PNG at 4096 × 4096 · matched, literally
  • "normal maps … intact" — Meshy file has no normal map · not matched
  • "silhouette intact" — I did not measure silhouette, so I do not score it

Meshy's viewer on the mobile version: Faces 4,962.

The triangle claim deserves full credit. Meshy's viewer said 4,962, the file said 4,962, and the job ran with 0 failures and 0 retries. Meshy's remesh is not the weak point of this chest. The reply is.

Two textures where the original had three

The Meshy 3D model I approved in message 2 was a 91.4 MB GLB with 2,052,398 triangles and three textures, a normal map among them. The approval card set the texture size to 4096 for 30 credits.

Meshy image-to-3D approval card: Meshy 7.1, Standard, texture 4096, Est. Cost 30.

My offline render of Meshy's 3D model (2,052,398 triangles, 91.4 MB).

The Meshy mobile GLB kept two of those three textures and dropped the third:

  • Base color — Meshy mobile 4096 × 4096, 22.5 MB
  • Metal/roughness — Meshy mobile 4096 × 4096, 14.2 MB
  • Normal map — Meshy original yes · Meshy mobile none
  • Texture pixels — Meshy mobile 33,554,432 · Snapshapy mobile 3,145,728 · 10.7×

I want to be precise about the wording: the Meshy mobile file has no normal map at all. Anyone who read the reply would expect one, and the only way to learn otherwise was to open the GLB.

My render of Meshy's downloaded mobile GLB (37.0 MB).

"Full texture fidelity" is the half of the reply that held, and for a phone it is the costly half: both maps stayed at the full 4096 size. The reply never said the textures would shrink, so I do not count that as a broken promise. I count it against "lightweight".

Wireframe close-up of Meshy's 4,962-triangle mobile mesh.

Snapshapy hedged before the remesh ran

Snapshapy's reply to message 3 came before the job finished, and it set expectations in writing: "A mobile-ready remesh is running now (free, ~30 seconds): about 5,000 triangles with textures kept and shrunk to 1024 px. UV-aware decimation can stop short of a low target on seam-heavy models, so I'll report the real triangle count when it lands."

Snapshapy message 3: "about 5,000 triangles … shrunk to 1024 px"; card "8.7k triangles · seams stopped it above 5.0k".

That reply contains five checkable claims: free, about 30 seconds, about 5,000 triangles, textures shrunk to 1024 px, and a real count to follow. The warning is the part I value most. It told me which number might not hold before I had a file to check.

"Seams stopped it above 5.0k": the card admitted the miss

The job finished in 12.1 s, under its own ~30-second estimate, and cost 0 credits. The card read "8.7k triangles · from 1.5M · textures kept · seams stopped it above 5.0k". The viewer showed 8,654 triangles and 12,414 vertices, one chest.

Snapshapy's viewer on the mobile version: 8,654 triangles.

Meshy's mobile mesh is the lighter one, 4,962 triangles against 8,654, and Snapshapy missed its own 5,000 target by 1.73×.

That miss is real and I do not soften it: 3,654 triangles over the number Snapshapy itself named. But the miss reached me through the card, in the same view as the result, not through my own file inspection afterwards. On the Meshy side, the one inaccurate claim reached me only because I opened the GLB.

Wireframe close-up of Snapshapy's 8,654-triangle mobile mesh.

Snapshapy's file against its reply: three 1024 maps

The textures matched the reply exactly. Snapshapy's mobile GLB holds three PNG textures at 1024 × 1024:

  • Base color — Snapshapy 1024 × 1024, 1.07 MB
  • Normal map — Meshy none · Snapshapy 1024 × 1024, 2.43 MB
  • Metal/roughness — Snapshapy 1024 × 1024, 0.60 MB
  • All textures — Meshy 36.6 MB · Snapshapy 4.10 MB

The normal map is the largest of the three, 59 % of Snapshapy's texture weight, and it is the one Meshy's file lacks.

My render of Snapshapy's exported mobile GLB (4.55 MB, three 1024 textures).

As an aside on weight: Meshy's mobile GLB is 37.0 MB against 4.55 MB, 8.1× larger, about 24.7 s against 3.0 s to download on 12 Mbps. Getting the Snapshapy file took 2 clicks, Export then GLB, though the in-browser GLB export of the mobile file took 52.4 s.

The earlier replies held up on both sides

Before message 3, both agents told the truth about what they were doing.

Meshy's first approval card said "4 paid actions need your approval", 9 credits each, Est. Total Cost 36; the balance went 856 → 820. Its second card said Est. Cost 30; the balance went 820 → 790. Meshy's four concepts were genuinely different chests (Dwarven Runic, Paladin Arcane, Deep Sea Coral, Mimic Monster), and its 3D model matched concept 1 closely.

Meshy's four concepts: Dwarven Runic, Paladin Arcane, Deep Sea Coral, Mimic Monster.

Snapshapy's reply to message 1 arrived within 9 s: "Four chest concepts are drawing now (8 credits)", naming an oak-and-iron classic, an ivory royal treasury chest, a mossy shipwreck chest and a rune-etched hex chest. Four single, clearly different chests were ready in about 43 s, against about 310 s for Meshy after approving.

Snapshapy's four concepts: oak-and-iron with coins, ivory royal, mossy shipwreck, violet rune hex chest.

Message 2 on Snapshapy said "3 credits, 5–8 minutes" and the model was ready after 335 s, about 5.6 minutes and inside its own range, though Meshy's image-to-3D step was faster at about 139 s. The header then read "11 spent", which is 8 + 3.

Snapshapy message 2: image-to-3D card ready, 3 credits, header "11 spent".

So the difference does not lie in messages 1 and 2. Every cost both agents stated matched the charge. It lies in the one reply that describes a file.

Every claim, checked against what I got

Here is the whole test reduced to statements and outcomes:

  • Triangle count — Meshy said 4,962, got 4,962 · Snapshapy said about 5,000, got 8,654 and flagged it
  • Textures — Meshy said full fidelity, got 2 maps at 4096 · Snapshapy said shrunk to 1024 px, got 3 maps at 1024
  • Normal map — Meshy said kept, got none · Snapshapy said textures kept, got a 1024 normal map
  • Warning given in advance — Meshy none · Snapshapy "can stop short of a low target"
  • Remesh time — Meshy about 117 s to the "ready" reply · Snapshapy said ~30 s, took 12.1 s
  • Stated credits — Meshy 36 + 30 + 0 = 66, matched · Snapshapy 8 + 3 + 0 = 11, matched
  • Total job time — Meshy about 566 s · Snapshapy about 390 s

Each agent got one number wrong in its message 3 reply. Meshy's wrong claim concerned a texture the file does not have, and it gave no hint. Snapshapy's wrong number was the one it warned about, and its own card corrected it.

Meshy after the run: balance 790 (started at 856).

On credits, as a side note: Meshy's balance fell from 856 to 790, 66 credits or $1.32 at $0.02 a credit, with 2 Approve clicks under the default "Always Ask". Snapshapy charged 11 credits, $0.52 at Starter rates, with 0 approvals; a ~$20 plan buys 15 Meshy chests a month against 36 on Snapshapy Starter. The pricing article in this series covers that bill.

Final verdict: Meshy 3D Agent scores 55/100 for a mobile game prop

For a mobile prop, the reply is part of the deliverable. Meshy 3D Agent told me the chest kept its normal maps, and the 37.0 MB file it gave me has none. It kept both remaining textures at 4096 × 4096, 33,554,432 pixels, and said nothing about the weight. Ten chests on this pattern would mean 660 credits, $13.20, 20 approvals, 370 MB of mobile files, and ten GLBs I would have to open to check which maps made it.

Meshy earns real points: an exact 4,962-triangle remesh, cost estimates that matched every charge, 0 failures, four genuinely different concepts, and the faster image-to-3D step. Snapshapy's regrets are the missed triangle target and the wait for its in-browser export. What decided my score is that Snapshapy's card reported its own miss, while Meshy's reply described a map its file lacks.

Which one I would use for this job

For a mobile game prop, I would use Snapshapy. Its mobile chest arrived with base color, normal and metal/roughness maps at 1024 × 1024, a card that printed the real triangle count, 11 credits and no approval stops. The plans are at snapshapy.com.

If a hard cap of about 5,000 triangles matters more than anything else and you are ready to shrink textures and supply a normal map yourself, Meshy Pro at $20 hits that count, at 66 credits per chest.