Day 2: Lights on, a bigger map and twelve characters


TL;DR: Friday started with the black screen still unsolved and ended with a lit, much bigger Segment 1, twelve characters modelled and imported, and Codex chewing through a queue of nineteen tickets. In between, Claude hit its usage limit twice, I had to admit the level was far too small, and Tripo taught me the hard way which side is left.
Morning: chasing the black screen
I woke up to the same black screen I’d gone to bed with. Our first suspect was something new in Unreal 5.9, so Claude sent a research thread digging. It found a real problem, just not that one: UE 5.9 starts its own MCP server on port 8000, the same port our editor connection used, which explained why Claude’s link to the editor kept dropping. (The fix is to start the game on a different port when the editor is also open.) MegaLights, the new lighting system, was the next suspect.
Then I did the simplest test of all. I opened the Segment 1 map in the editor’s own viewport, and it was black there too. That meant the problem wasn’t the camera or the game: it was the map itself. Within a quarter of an hour Claude had it. The map had no lights at all, and the test sun we’d added was pointing at the sky, because Unreal’s Python rotator takes its angles in a different order from the one Claude assumed. One fix later, at 9:07 AM, Segment 1 rendered lit for the first time (PR #31). It was a humbling bug. We’d been ready to blame a brand-new engine, and it was a missing light switch.
Voices for the tavern crew
While the level was dark, the audio wasn’t. Claude planned the Segment 1 audio and auditioned voices in ElevenLabs using real lines from the script. My verdict on the first round was that they felt flat. For round two, Claude wrote each line inside a short scene with audio tags for breath, pace and mood, and the difference was night and day. Five voices are saved now: a West Country Tavernkeeper, a gruff Southern English Guard, a husky Yorkshire Hunter, a Glaswegian Brawler and a rasping, ancient Briar Warden.
I also asked for a launch trailer as a separate project. It isn’t the jam video, which has to be 30 to 60 seconds of real gameplay, so it can take its time. I chose a serious, creepy story with no narrator: the heroes run out of mead in the castle tavern and set off for the village. And I set a rule for the rest of the jam: Unreal and scene work runs on Claude Opus 5.5 at medium effort.
Midday: real assets, and “this is too small”
With the lights on, it was time to move from greybox to real assets. With my OK, Claude used computer use to open the Unreal editor itself and installed the first Megascans through the editor’s Fab tools: a granite rock and a forest-floor ground (PR #32). The first pass along the river path looked rough, but it was a real material on real ground, and that felt like progress.

The first Megascans boulders along the river path, in the original small map.
Then I opened the level and said what I’d been thinking: it was a 30-second straight line. I was specific about what I wanted, and it’s still the brief for the whole map. Think of England in the 1700s. There were no straight roads and no flat ground. Paths should wind, players should be able to get a little lost but not too lost, and there should be more to explore.
Claude proposed a main path of about 1 km winding down from the castle hill and splitting into three routes. I added two rules that also help performance: leave room on the farm for wheat fields and paddocks, and don’t dress the woods between routes. Like the classic co-op shooters, the routes are fixed corridors, and only what you can see from the path gets detail. By 12:30 PM the bigger Segment 1 was built: hilly terrain, three winding routes about 250 m apart, and about five minutes of walking before you add a single fight.

The second route plan. Three ways through, each with its own landmarks.

Concept art for the courtyard, which I generated in ChatGPT from Claude’s prompts. The level is built to match its mood, not copied from it.
Claude also wrote two Cursor tasks as GitHub issues, each ending in a prompt I can paste: a themed 404 page for this site and a balance-table checker for the game.
Afternoon: twelve characters through Tripo and Blender
This was the day the cast came together. I generated turnaround sheets in ChatGPT, built the meshes in Tripo Studio with Smart Mesh, and exported them without a rig. Claude then took each one into Blender, cleaned it up, set its real height and fitted it to our shared skeleton so they can all use the same animations.
The Briar Warden went first: 1.9 m tall, about 17,700 triangles, with the robe moving cleanly in a stride test (PR #33). The heroes followed at their correct heights (PR #49), then the Thrall and the Thorn Hulking (PR #56), and by the evening the five specials (PR #64).

The heroes after Tripo and Blender.

The Guard, checked from every side before going into Unreal.

The Briar Warden, rigged. The first character in the game.

The Thrall, the common enemy you’ll meet in packs.

The Thorn Hulking, the heavy that ends Segment 1.

The Caskbound, one of the five specials.

The Rookcaller’s rig, with its ragged wings spread.
Codex at full speed, and new UI art
With Claude on the level and the characters, I wanted Codex carrying most of the code. Claude wrote nineteen detailed Codex tickets, in order: mead and Ruckus, class fit, difficulty, route blockers, bots, objectives, the pressure director, enemy AI, pings, the HUD, then each monster. I set the bot rules: stay within about 50 m of the player, recover when stuck, and fight, defend and revive like real teammates without playing perfectly. I also came up with one of my favourite ideas of the jam: fields full of T-posing scarecrows, any one of which might be the real Strawman.
Codex’s first big batch landed after passing its tests: the UI framework, the Tavernkeeper’s Cask Spill, mead drinking with the Ruckus meter, and the hidden class-fit maths (PRs #30, #57, #61 and #62).
I also generated a set of medieval UI tiles in ChatGPT: carved signs, iron frames, a crimson banner, gold scrollwork and the Left for Mead logo with clinking tankards. Claude keyed out the backgrounds, cropped them into 26 textures and rebuilt the menu mockups with them. That’s when we decided the real screens would be built in C++ rather than in the visual editor, so Codex could build and test them without a screen. The same art dresses this website.
Evening: the second pass
The second Segment 1 pass went into main (PR #35): real spruces lining the paths, cobblestone roads, a bigger farm with wheat plots and paddocks, and eleven scarecrows. I asked for concept art of every Segment 1 area so the level could be built towards it. Claude wrote the prompts, and I generated thirteen images in about twenty minutes. A gear plan followed: most weapons come free from Fab, but Fab armour won’t fit my Tripo heroes, so each hero gets mail and plate versions that I’ll make in Tripo.

The farm after the second pass, captured early the next morning. Real trees and cobbles, still a long way from finished.
My honest estimate at the end of the day was about 30 to 35% of a finished Segment 1. The level and hero gameplay were furthest along. Enemies, the director and bots were the biggest risk.
The hard parts
Usage limits. Claude hit its five-hour usage limit twice, at 9:45 AM and again at 1:15 PM, and work stopped for a few hours each time. In a four-day jam, that hurts.
Merges blocked on my PC. The safety check on my PC blocked merges and force pushes even when I’d approved them, because the approval reached it through Claude rather than in my own words. Merging moved to cloud threads for good.
Which side is left? My first Smart Mesh run on the Guard came out with left and right swapped. Tripo’s “left” view means the character’s own left, not the viewer’s, and you must never mirror an asymmetric character to fill in a missing side. I re-ran it, and the rule went into our notes.
Large files from the cloud. Cloud threads can’t push Git LFS files, so the UI art had to wait to be finished from my PC.
Disk space. Old build copies were eating my main drive. A clean-up freed about 15.6 GB, which bought me room to keep building.
Tools I used, and for what
Claude Code Desktop on my PC ran the Unreal work, the Blender clean-ups and the imports, and Claude computer use drove the editor directly when scripts couldn’t. Codex CLI worked through the ticket queue. Tripo turned my ChatGPT turnarounds into meshes, Blender and Mixamo got them rigged, and ElevenLabs gave the cast their voices. Fab and Megascans supplied the rocks, ground and trees. Cursor took two small tasks, and Hyper3D Rodin was lined up for the props Fab doesn’t have.
