bunpav
← Back to stories

bunpav ·

The Game Jam 3D Asset Pipeline: Generate, Rig, and Export in Under 48 Hours

A practical game jam 3D assets pipeline for Ludum Dare and Global Game Jam teams: generate a batch fast, lock style once, auto-rig once, export once.

12 min readbunpav crewAIGame Dev Tools3D AssetsIndie gamesText-to-3D

If you've run a 48-hour game jam before, you know the ugly secret: art gets whatever time is left after code and design take theirs, which in practice means a handful of exhausted hours at the end. Solving that is the whole point of a real game jam 3D assets pipeline — not "make better art," but "make enough art, fast enough, that it stops sinking your submission." Ludum Dare, Global Game Jam, and most itch.io jams run on the same brutal clock, and teams that ship a finished-looking game almost always treated 3D assets as a scheduling problem first and an artistic one second.

This is a practical, hour-by-hour production guide for exactly that: when to generate, when to lock your style and stop generating, how auto-rigging and a single export pass save you from the two most common late-jam disasters, and what to skip entirely so your last hours go to bugs and juice instead of topology.

TL;DR — the questions people actually ask

QuestionDirect answer
How much jam time should go to 3D assets?6-10 hours total, front-loaded into one big generation batch, not spread evenly across the whole jam.
When should I lock my art style?By hour 2-3, before any asset generation starts — changing style mid-jam is the single biggest time-waster.
Do I need to rig characters by hand?No — auto-rig every humanoid or creature mesh in one batch, once, right after generation.
How many times should I export?Once per asset, ideally by hour 30-34, with a hard stop on new geometry after that.
Can I legally use AI-generated assets in Ludum Dare or GGJ?Yes — both jams have published policies that explicitly allow it (see below), though category opt-outs can apply.
What kills game jam submissions visually?One over-polished hero asset next to placeholder cubes — not rough art applied consistently.

Why does 3D art always run out of time first in a game jam?

Game jams run on a hard external deadline, and every hour one discipline gets is an hour another doesn't. Ludum Dare's rules define the Compo as a 48-hour solo challenge where "all your content — art, music, sound — must be created in 48 hours," while the Jam track relaxes that to 72 hours with teams allowed, per the official Ludum Dare rules on ldjam.com. Global Game Jam runs its own fixed 48-hour window at every site worldwide, theme revealed at 5 p.m. local time Friday, submissions due 48 hours later, per globalgamejam.org.

Neither clock cares whether your art is done. Code needs a playable state before anyone can judge the game at all, design needs the core loop locked before art knows what shape assets should take, and testing has to happen before submission — so art is structurally squeezed from both sides. A game jam survival guide from Gamedō puts the fix bluntly: cut your first idea in half, then in half again, because "one mechanic, one level, one enemy type" is usually still ambitious for 48 hours — and that scope discipline has to cover art asset count too, not just gameplay features.

The fix isn't working faster by hand. It's compressing the slowest part of the traditional pipeline — blocking geometry, UVs, a texture pass, a skeleton — into a generation step that takes minutes, so human time goes to the last-mile calls a generator can't make.

How do Ludum Dare and Global Game Jam actually structure their clocks?

Ludum Dare: Compo vs. Jam, and an extra hour you shouldn't waste

Ludum Dare splits into two concurrent tracks. The Compo is solo, 48 hours, and every asset — including art — must be original work made within that window. The Jam extends to 72 hours, allows teams, and permits third-party or pre-made assets, though entrants using them are expected to opt out of the Graphics and Audio judging categories, per Ludum Dare's rules page. Both tracks add a one-hour Submission Hour after the timer ends purely for packaging and uploading, not for finishing assets — a distinction plenty of teams learn the hard way.

Ludum Dare has also published a direct answer on generative tools: its FAQ states AI-powered tools can be used, but jammers "may have to opt out of certain categories" where a generator did most of the creative work — see Ludum Dare's "Can I use AI?" page. That's meaningfully different from a blanket ban, and it matters for anyone treating generation as part of their normal pipeline rather than a shortcut around the rules.

Global Game Jam: one weekend, one theme, no do-overs

Global Game Jam runs a fixed 48-hour window at every registered site, synced to a secret theme revealed simultaneously worldwide at the start of the event. The first GGJ in 2009 drew roughly 1,600 participants across 23 countries producing 307 games, and it has scaled into one of the largest simultaneous game-making events on the planet since, per globalgamejam.org and Wikipedia's coverage. GGJ has also published a standalone AI policy stating it "will not have any restrictions around AI usage within the jam, including generative AI," so long as output respects the jam's existing copyright and IP rules — see globalgamejam.org's AI policy announcement.

Neither event will disqualify you for generating a mesh from a prompt, and neither gives you extra time for it — which is the actual argument for a tight pipeline instead of hoping for the best.

What does a realistic game jam 3D asset pipeline look like hour by hour?

Treat this as a template, not a rulebook — adjust the hours to your team's engine familiarity, and to whether you're running a 48-hour Compo/GGJ format or a 72-hour Jam. The shape stays the same: one early batch, one style lock, one rig pass, one export, then stop touching geometry.

HoursArt/asset phaseWhat happens
0-2Theme reveal + scope lockLock the core loop and a hard cap on asset count — every prop/character on the list gets built once, no "add more later."
2-3Style lockOne reference image, one palette, one poly-density target — the single most important 60 minutes for art, since changing it later multiplies rework.
3-8Batch generationGenerate every major prop and character in one sitting, picking the cleanest silhouette from a few quick rerolls rather than perfecting one prompt.
8-11Auto-rig passRun every humanoid or creature mesh through auto-rigging in one batch — skeleton and skin weights, not manual weight painting.
11-13First export + import testExport the batch once (GLB by default) into the engine to catch pivot, scale, and material issues while there's time to regenerate.
13-30Core buildTeam builds the actual game against real assets instead of placeholder cubes; art does spot texture/recolor tweaks only, not new geometry.
30-34Last generation windowOne final pass for genuine gaps — a missing enemy variant, a UI prop — and nothing else. Hard cutoff for new asset creation.
34-44Polish + juiceLighting, particles, sound, screen shake, UI — applied evenly across every asset rather than piled onto one hero prop.
44-47Build lockFinal export, a clean build on the submission platform, and a full playthrough by someone fresh to the project.
47-48 (+ Submission Hour)Ship itPackage, write the description, upload. No new art, no "one more tweak."

A 72-hour Ludum Dare Jam or a multi-day GGJ site just stretches these same phases proportionally — the ratio of "one batch, one lock, one export" matters more than the absolute hour count.

What should you not do with your art time in a jam?

Three failure patterns show up in almost every rushed jam postmortem, and all three are avoidable with the pipeline above.

Chasing perfect topology. Nobody judging a 48-hour game is inspecting your edge flow. Retopology and manual UV unwrapping are "invisible quality" that eats hours for zero visible payoff — save that discipline for a post-jam version if the game is worth continuing.

Manual rigging from scratch. Placing joints, painting skin weights, and testing deformation by hand is a real craft, and real crafts take real time even for professionals. A jam has no slack for it — auto-rigging exists specifically to compress this into seconds so a solo Compo entrant isn't stuck animating a T-pose at hour 40.

Over-polishing one hero asset. itch.io's guide to game jam art argues for "quality over quantity" applied to reused assets specifically — whatever the player sees most often deserves the extra pass, not whichever prop the artist personally likes best. A gorgeous main character next to bare gray placeholder crates reads as an unfinished game, even if the character alone would win an art contest.

What's the export/import checklist for whatever engine your team is using?

Run this once, right after batch generation and the rig pass — not asset-by-asset as an afterthought:

  1. Pick one export format for the whole batch. GLB is the safe default for Unity, Unreal, and Godot alike — a single self-contained glTF file with mesh, materials, and textures bundled together, no separate texture-linking step to fumble at 2 a.m.
  2. Import the first two or three assets immediately, not the whole batch at once — catch scale, pivot, and material issues on a small sample before they're baked into twenty more.
  3. Check scale and pivot against your engine's unit conventions — a mesh at the wrong real-world scale breaks physics, camera framing, and UI overlap in ways that are annoying to debug blind.
  4. Verify rigged characters actually animate with at least a placeholder clip before committing to that skeleton — a rig that looks correct in a viewer can still deform oddly under your engine's specific animation system.
  5. Lock the export and stop regenerating once the batch is in and tested. Re-exporting "one more tweak" late in a jam is a disproportionately common source of last-hour build breaks.

If your team is torn between formats or hits engine-specific import quirks, the deeper breakdowns on GLB vs. FBX vs. OBJ vs. USDZ and what actually survives the Unity/Unreal import cover the failure modes in more depth than a jam weekend allows — worth a skim before the jam starts, not during it.

What should you do — a jam-day asset checklist?

  1. Lock your art style in hour 2-3, before anyone generates a single asset — one reference image, one palette, one poly-density target, written where the whole team can see it.
  2. Batch-generate the entire prop and character list in one sitting, picking the cleanest silhouette from a couple of quick rerolls per asset instead of perfecting one prompt.
  3. Auto-rig every humanoid or creature mesh in one pass, immediately after generation — never mid-build, never one at a time.
  4. Export once, early enough to test the import — around hour 11-13 on a 48-hour clock — so scale, pivot, and material problems surface while there's time to fix them.
  5. Set a hard cutoff for new geometry around hour 30-34, and spend everything after that on juice, polish, and bug fixing.
  6. Polish evenly, not heroically — a consistent pass across everything on screen beats one spectacular asset next to placeholders.
  7. Do a full playthrough before submission with someone who hasn't stared at the build for 40 hours — they'll catch the broken import everyone else stopped seeing.

How does bunpav fit into a jam-weekend pipeline?

bunpav turns a text prompt or a reference photo into a textured, game-ready mesh, with optional auto-rigging for humanoid or creature skeletons and export to GLB, FBX, OBJ, or USDZ — mapping directly onto the "batch generate, rig once, export once" shape above. It's currently in private beta on a waitlist, and pricing when it opens runs on prepaid credit packs instead of a subscription: Starter at $9 for 100 credits, Studio at $39 for 550 credits (sized, in our own framing, for a jam weekend or a small prop library), and Pro at $89 for 1,500 credits. Details on the pricing section.

We won't pretend generation removes every risk — non-determinism means you'll still reroll prompts, and hands, hinges, and legible on-surface text remain weak spots industry-wide, covered in the text-to-3D workflow guide. What it does is collapse the part of the pipeline with no business eating jam hours — blocking geometry, base UVs, a texture pass, a skeleton — so jam time goes to the game, not the mesh.

If your jam game leans into chaotic multiplayer instead of solo scope, Top 10 Friendslop Games in 2026 is a good reference point for how much the "held together by proximity chat" genre rewards a simple, evenly-polished asset set over one showpiece character.


Jam formats, rules, and AI policies described here reflect Ludum Dare and Global Game Jam's published pages as of July 17, 2026. Jam rules and platform policies change between events — check ldjam.com and globalgamejam.org directly before relying on specifics here for an active submission.

Player questions

How much time should 3D art actually get in a 48-hour game jam?

Roughly 6 to 10 hours of hands-on asset work, split into one generation batch early and one cleanup pass late — not a continuous grind. Manual modeling and rigging routinely eat 15-20 hours for the same asset count, which is exactly the time a jam team doesn't have once code, design, and testing take their share.

Can you use AI-generated 3D models in Ludum Dare or Global Game Jam?

Yes, both events explicitly allow it. Ludum Dare's policy says AI tools can be used but you may need to opt out of judging categories where generation did most of the work, and Global Game Jam's AI policy places no restriction on generative AI beyond its existing copyright and IP rules.

What's the biggest 3D asset mistake teams make in a game jam?

Over-polishing one hero asset while the rest of the game ships with placeholder cubes. A jam is judged on the whole 90-second impression, so an evenly-finished game with simpler assets consistently reads better than one gorgeous character next to bare gray boxes.

Should you rig characters by hand in a 48-hour jam?

No — manual rigging and weight painting is a multi-hour skill even for an experienced technical artist, and a jam has no slack for it. Auto-rigging tools place a skeleton and skin weights on a humanoid or creature mesh in seconds, which is the only version of rigging that fits a jam's clock.

What file format should game jam 3D assets export to?

GLB for Unity, Unreal, Godot, and most web/WebGL jam builds — it's a single self-contained file with mesh, materials, and textures. Reach for FBX only if your engine needs separate animation data, and USDZ or OBJ if a judge or teammate needs to preview the model outside the engine.

Is bunpav actually useful for a game jam, or is it overkill?

It's built for exactly this time pressure — prompt or photo in, textured mesh with optional auto-rig out, exported to GLB, FBX, OBJ, or USDZ. bunpav is in private beta with a waitlist, and the Studio credit pack ($39 for 550 credits) is sized around what the team calls a jam weekend rather than months of ongoing production.

More to read