How to Try Primitive Lab - First Indie Procedural GLB Spike 2026

If you are searching primitive lab, you do not need another “AI 3D will replace artists” thread. You need a Monday path: sign into bunpav’s browser studio, pick one prop or platform, lock a seed, export a pivot-correct GLB, drop it into Unity / Unreal / three.js (or Blender), then decide keep / hold / rewrite before anyone burns Meshy credits on every crate in the jam level.

On July 19, 2026, bunpav shipped Primitive Lab — a browser-based procedural (not diffusion) asset studio at /dashboard/studio. The feature page and product surface describe parametric sliders, deterministic seeds, live materials, and one-click binary GLB (plus a copy-paste GLTFLoader snippet). During beta, procedural builds are free for signed-in users; prepaid credit packs still meter separate text-to-3D / image-to-3D features.

This evening is not Meshy 8K Texture (AI albedo). It is not Material Maker 1.7 (node PBR graphs). It is not AutoRemesher (topology). It is a primitive lab keep/hold for jam blockout props that must rebuild identically tomorrow.

Time: about 60–90 minutes for sign-in + one asset + one engine import. Longer if you batch a whole set dressing pass the same night (do not — finish smoke first).

Why this matters now

Three clocks make primitive lab worth an evening this month:

1. A dated product ship, not a vague “procedural tools exist” claim. bunpav’s Jul 19 blog positions Primitive Lab as distinct from their own AI mesh pipelines — the Discord argument you will hear is “is this just another Meshy clone?” Your receipt must answer no with a seed rebuild.
2. Jam and web-game art time is still the bottleneck. Parent three.js demand stays enormous (~114K GitHub stars; tens of millions of monthly npm downloads for three). Browser-native, pivot-correct GLB sits exactly where PROCJAM-style and Gamedev.js teams already work.
3. Credit economics changed the default instinct. Teams that spent June–July re-rolling AI meshes now need a free blockout lane. Primitive Lab’s beta story is explicit: procedural studio does not spend the same credit packs as text/image-to-3D (pricing).

Already used: Meshy 8K, Material Maker 1.7, Meshy 3D Agent. Add BUILD_RECEIPT if your studio files art spikes beside ship evidence, and AI 3D import checklist when heroes graduate to generative meshes.

Glossary - plain words

What Primitive Lab actually ships (true summary)

From the official blog, feature page, and product.md (verify live UI — library counts grow):

• Browser studio on three.js + React Three Fiber — no local DCC install for the spike.
• Parametric library — vehicles, rocks, props, platforms at launch; feature page now cites a larger multi-category registry (animals, trees, characters, nature — confirm in-product before claiming inventory in a partner deck).
• Geometry sliders per asset (length, jaggedness, tire radius, etc.) + overall scale + variation seed.
• Live materials — Toon / Lambert / PBR / Gloss + built-in palettes + custom hex.
• Deterministic rebuild — same asset ID + seed + parameters → identical geometry (mulberry32 PRNG story on the feature page).
• Export — one-click GLB; optional “Use in project” GLTFLoader snippet for three.js / R3F.
• Pivots — centered origin, rests on y = 0, dimensions in metres; live tri/mesh counts in viewport.
• Beta pricing — Primitive Lab free for signed-in users; AI text/image-to-3D still on credit packs (Starter / Studio / Pro tiers on pricing — re-check before budgeting).
• Optional image→procedural sculpt (product.md notes a credit-cost path) — out of scope for this first evening; stay on pure slider smoke.

Keep / hold / rewrite - Monday decisions

Beginner path - first prop in one evening

Prerequisites

• Modern desktop browser (Chrome/Edge/Firefox current).
• Ability to create/sign into a bunpav account (beta flow — follow on-site prompts).
• One target engine or a three.js sandbox (Unity, Unreal, Godot, Blender, or a tiny R3F page).
• 60–90 minutes without “also generate the whole city with AI.”

Steps (first path)

1. Open the studio. Sign in and go to bunpav dashboard studio (path cited in the blog). Confirm you are in Primitive Lab, not Audio Lab / Game Lab / AI generate.
2. Pick a boring win. Choose a crate, barrel, or ramp/platform — not the fanciest vehicle on night one.
3. Nudge two sliders only. Change overall scale and one shape control. Watch the mesh rebuild live. Stop when it reads as “good enough blockout.”
4. Lock the seed. Copy the seed (and note asset ID + slider values) into a Discord note or seeds.txt. Same inputs must rebuild tomorrow.
5. Try “prompt a variation” once. Type a short word, observe seed/palette nudge, type the same word again — confirm it repeats. This is deterministic mapping, not ChatGPT mesh magic.
6. Export GLB. Download the binary glTF. Optionally copy the GLTFLoader snippet if you are on three.js.
7. Import once. Drop into your engine or Blender. Check: sits on floor, origin at center, scale roughly metres, materials readable under your lighting.
8. Write keep/hold. One sentence: Keep for jam props / Hold until beta TOS clear / Rewrite — need unique hero.

Beginner success screenshot checklist

• Studio viewport with asset + visible seed/params (redact account email).
• Engine or three.js view with the mesh on a ground plane.
• Side-by-side note: “rebuilt from seed X” after a second export (optional stretch).

Discord brief you can paste

Primitive Lab spike (60–90m):
- Sign in → /dashboard/studio → ONE prop/platform only
- Lock seed + asset ID + slider values in seeds.txt
- Export GLB → import Unity/Unreal/three.js/Blender
- Confirm pivot on y=0 + metre scale
- Decision: keep / hold / rewrite (procedural ≠ Meshy)
- No AI credit burn on this pass; heroes stay separate evening

Developer gates (P1–P6)

Gate deep-dive

P1 — Access. Beta products change invite walls. If you cannot enter the studio, mark hold — do not invent a “we tried” narrative for diligence. Document the block (waitlist, region, SSO failure).

P2 — Parametric smoke. Prove the tool is procedural: dragging a slider must rebuild without a generation queue. If every change spins a credit meter, you are in the wrong bunpav product surface.

P3 — Seed lock. This is the production property. Store asset_id, seed, slider map, material mode, palette name/hex. Treat it like a content address. Teammates should rebuild without the GLB if needed.

P4 — Export. Prefer GLB over screenshot tourism. If export fails, stop — do not screenshot-only “pass” the evening.

P5 — Engine honesty. Unity ≈ metres; Unreal often thinks in centimetres — if the crate is a skyscraper, fix import scale and write it on the receipt. Confirm materials are not pure magenta/error. Visual mesh ≠ physics collider — note that you still need collision volumes.

P6 — Keep/hold. Company readers care whether “procedural beta” is allowed in shipped builds and store pages. Pair with AI disclosure policy if any other assets on the page are generative (Steam AI disclosure debate context when relevant to your store plan).

Receipt template

{
  "receipt_type": "primitive_lab_glb_receipt_v1",
  "date": "2026-07-27",
  "operator": "your_handle",
  "studio_url": "https://bunpav.com/dashboard/studio",
  "product": "Primitive Lab",
  "beta_status": "signed_in_free_procedural",
  "asset_id": "crate_or_platform_id",
  "seed": "123456789",
  "sliders": {
    "scale": 1.0,
    "shape_control": "as_set"
  },
  "material_mode": "PBR",
  "palette": "built_in_or_custom_hex",
  "export_format": "glb",
  "file_name": "jam_crate_seed123.glb",
  "import_targets": ["unity|unreal|godot|blender|threejs"],
  "pivot_ok": true,
  "metre_scale_ok": true,
  "collider_note": "visual_only_author_in_engine",
  "credits_spent": 0,
  "decision": "keep|hold|rewrite",
  "procedural_vs_ai_sentence": "Blockout props from Primitive Lab seeds; heroes via separate AI/DCC evening.",
  "gates": {
    "P1": "pass|fail",
    "P2": "pass|fail",
    "P3": "pass|fail",
    "P4": "pass|fail",
    "P5": "pass|fail",
    "P6": "pass|fail"
  },
  "notes": ""
}

File beside screenshots. If any gate is fail, decision cannot be keep for production paths.

Procedural vs AI - decision table (cite-worthy)

Use this table in stand-ups. The point of primitive lab is not “never use AI” — it is stop paying diffusion tax for every barrel.

Engine import notes

Unity

• Import GLB/glTF with a supported importer (UnityGLTF or package of record for your LTS).
• Expect 1 unit ≈ 1 metre if Primitive Lab metre scale held.
• Add Box/Mesh colliders; do not assume export includes gameplay collision.
• Capture a frame with default directional light — PBR roughness should read.

Unreal

• Import via glTF pipeline your project already trusts (or Interchange).
• Watch cm vs m — a “1 m crate” may need ×100 or a documented scale actor.
• Nanite optional; for jam props usually off.
• Screenshot with Lumen/MegaLights as your ship path — do not smoke only in unlit.

Godot 4.x

• Drag GLB into res://; instance in a throwaway scene.
• Check root node transform; apply if artists rotated in confusion.
• StandardMaterial3D should pick up base color; fix metallic/roughness if washed out.

three.js / React Three Fiber

• Paste the studio’s GLTFLoader snippet or load the GLB with your existing loader.
• This is the highest-fidelity “same stack” path — Primitive Lab is built on three.js/R3F per bunpav.
• Confirm position.y rests on your ground plane without manual offsets.

Blender

• File → Import → glTF 2.0.
• Useful for UV inspection and baking later — not required for the keep/hold smoke.

Common mistakes

1. Treating Primitive Lab as Meshy. Different category. If producers say “prompt the truck,” redirect to seed + sliders or open the AI product intentionally.
2. Skipping seed notes. Without seeds, you shipped a one-off GLB — you lost the procedural advantage.
3. Batching 30 assets night one. Finish P1–P6 on one mesh. Scale after the receipt is green.
4. Assuming free forever. Beta free ≠ forever free. Re-read pricing before roadmap commitments.
5. Calling visual mesh “game-ready physics.” Author colliders. Document on receipt.
6. Ignoring TOS / commercial terms. Confirm license for shipped games before Steam page copy claims.
7. Mixing credit-cost image→procedural sculpt into “free procedural” claims. product.md separates credit paths — label them honestly.
8. Hero albedo expectations. Palette modes ≠ native 8K Multiview Diffusion.

Honest limits

• Library coverage is finite — not “any object from a sentence.”
• Beta — features, categories, and pricing can move; verify live UI the week you ship.
• Materials are studio modes/palettes, not full Substance-class graphs.
• No substitute for animation-ready hero characters with production rigs (even if characters appear in the library — smoke carefully).
• Account wall — signed-in beta may block some teams; have a Kenney/CC0 fallback.
• Diligence: partners may still ask “where did art originate?” — seed receipts help.

Company / diligence notes

Put in the zip:

1. primitive_lab_glb_receipt_v1.json
2. Seed sheet (seeds.txt or CSV)
3. Import screenshots (Unity and/or Unreal if both are ship targets)
4. One-paragraph procedural vs generative policy
5. Link to bunpav blog + feature page dated for the spike
6. Confirmation that zero AI credits were spent or an itemized credit note if you also ran AI heroes

ROI framing: if a jam would have spent N credits re-rolling props, Primitive Lab’s free beta lane is the control experiment — measure hours saved, not vibes.

Schedule - 75-minute evening

Stretch goals (only after P6 pass)

• Rebuild the same asset from seed only (delete GLB, re-export) — prove determinism.
• Scatter three seed variants of one rock for a field — store seed list, not three mystery files.
• Import the same GLB into two engines; note scale deltas.
• Pair one procedural placeholder with one Meshy 8K hero in the same scene for a producer demo.

Key takeaways

1. Primitive Lab is bunpav’s procedural browser studio — sliders + seeds, not diffusion.
2. Jul 19, 2026 ship + feature page make this a timely keep/hold, not evergreen filler.
3. Free beta procedural builds ≠ free AI text-to-3D — keep credit packs separate in budgeting.
4. Lock asset ID + seed + sliders or you threw away reproducibility.
5. Export GLB and smoke pivot + metre scale in the real engine before batching.
6. Visual mesh ≠ collider — author physics in-engine.
7. Use Primitive Lab for coverage props; graduate heroes to AI/DCC intentionally.
8. Do not cannibalize Material Maker (graphs) or Meshy 8K (albedo).
9. File primitive_lab_glb_receipt_v1.json for diligence and Discord clarity.
10. Re-verify live library size and TOS the week you commit a ship page.
11. three.js/R3F teams get the cleanest drop-in via GLTFLoader snippets.
12. Keep/hold/rewrite is a written decision — screenshots alone are not a policy.

FAQ

What is Primitive Lab?

bunpav’s browser-based procedural 3D asset studio. You pick an asset, adjust geometry sliders and a seed, choose materials/palettes, and export a pivot-correct GLB — built on three.js / React Three Fiber (blog).

Is Primitive Lab AI-generated?

No for the core slider/seed path. It is parametric and deterministic. bunpav’s separate text-to-3D and image-to-3D tools are the generative AI products.

Is Primitive Lab free?

Free for signed-in users during beta for procedural studio use, per bunpav. AI generation still uses credit packs — confirm pricing before you promise producers “unlimited forever.”

What can I export?

Binary GLB (glTF) for Unity, Unreal, Godot, Blender, and three.js, plus an optional GLTFLoader code snippet for web projects.

Why does the seed matter?

Same seed + same parameters rebuild the identical mesh — shareable like code, useful for version control and teammate reproduction.

How is this different from Meshy or Tripo?

Those tools (and bunpav AI) generate novel meshes from prompts/photos — flexible but stochastic and usually metered. Primitive Lab trades “anything” for instant, free (beta), exact rebuilds across a fixed library.

How is this different from Material Maker?

Material Maker authors procedural material graphs (often Godot-based) for PBR export. Primitive Lab authors parametric meshes in the browser. Many pipelines can use both.

Can beginners finish in one evening?

Yes if scope is sign-in + one prop/platform + one import + receipt. No if you also try to fill the whole level and run AI heroes the same night.

Do exported meshes include colliders?

Expect visual geometry. Author collision in-engine and write that on the receipt.

What should studios put in diligence?

Receipt JSON, seed sheet, import screenshots, procedural-vs-AI policy sentence, and dated links to bunpav’s blog/feature pages.

Related reading

• How to Use Meshy 8K Texture - First Indie Hero Asset Evening 2026 — generative hero albedo (not procedural GLB)
• Material Maker 1.7 Steam Indie PBR Keep Hold Playbook 2026 — node PBR graphs (not browser mesh studio)
• How to Use Meshy 3D Agent - First Game-Ready Asset
• AI 3D Model Generator Tools - Import-Ready Checklist
• How to Use AutoRemesher Free - First Indie Quad Retopo Evening 2026
• How to Use Tripo DCC Bridge - First Unreal Unity Evening
• Your First BUILD_RECEIPT JSON
• Blender to Unity and Godot Export Pipeline

Official and useful sources

• Primitive Lab blog (Jul 19, 2026)
• Primitive Lab feature page
• bunpav product.md
• bunpav pricing
• bunpav home

Bottom line

primitive lab is a Jul 2026 browser procedural spike worth filing: seed-deterministic props, pivot-correct GLB, free beta lane beside metered AI. Prove one crate/platform in your real engine, lock the seed, write keep/hold, and refuse to spend diffusion credits on the junk drawer. That is the evening that survives Discord and partner diligence.