1 · Toolkit
$0
MIT-licensed skills, the um CLI, the knowledge base.
2 · Agent
Your plan
Claude Code, Codex, Gemini CLI, Copilot, Cursor or OpenCode: whatever you already pay for.
3 · Assets
~$0.5–7
fal, per generation, for a small mod's sprites, a 3D model and sound effects.
fal prices for the toolkit's default models
The um fal recipes use these models by default. Prices from fal's model pages, checked October 9, 2026.
| Asset | Default model | Command | Price |
|---|---|---|---|
| Sprite with transparency 1024×1024 | openai/gpt-image-2 | um fal sprite | $0.211 high (default) · $0.053 medium · $0.006 low |
| Image / key art / edit 2K and 4K cost more | fal-ai/nano-banana-2 | um fal image | $0.08 per image |
| 3D model from an image | fal-ai/trellis-2 | um fal model3d | $0.25 (512p) · $0.30 (1024p) · $0.35 (1536p) |
| Sound effect | fal-ai/elevenlabs/sound-effects/v2 | um fal sfx | $0.002 per second |
| Music, voice, video Open-ended; video is the most expensive | various | um fal music / voice / video | Check before batching |
| Images without fal Images only | local ComfyUI | um comfy image | $0 (your GPU) |
Estimate your mod
Change the counts to match your idea. Retries multiply the sprite cost; first attempts are rarely final.
Worked examples
| Mod | Assets | fal cost |
|---|---|---|
| Code-only mod (stats, rules, UI tweak) | none | $0 |
| One new weapon, test-quality art | 3 sprites, low | ~$0.02 |
| Small content pack | 10 sprites high, 1 image, 1 3D model, 10 s SFX | ~$2.51 |
| Same, three tries per sprite | 30 sprites high, 1 image, 1 3D model, 10 s SFX | ~$6.73 |
| AoE2-style unit from 3D | 1 concept image, 2 3D models (1536p), Blender renders free | ~$0.78 |
How to keep the bill down
- Placeholders first. Ask for fal art only after the mechanic works in game.
- Draft at low quality, then regenerate only the keepers at high.
- Price before you batch:
um fal price fal-ai/trellis-2. - Use ComfyUI locally for images if you have a GPU:
um comfy image --sprite. - Set a fal spending limit in the dashboard.
- Reuse the manifest.
um fallogs every generation, so nothing is regenerated by accident.
um fal price openai/gpt-image-2
um fal price fal-ai/trellis-2
um fal search "music" # find a model, then price itThe agent bill is separate
fal's dashboard doesn't show your agent usage. A long reverse-engineering session uses far more agent tokens than any asset batch costs. Use recon-first prompts, keep MODLOG.md so new sessions don't redo work, and let the agent start from the knowledge base instead of rediscovering gotchas.
Frequently asked questions
Is Universal Modder free?
Yes. The code is MIT licensed and free to use, modify and redistribute. What costs money is the AI agent you run it with and, optionally, fal for generated assets.
Do I have to pay for fal?
Only if you want generated art, 3D, audio or video. Code-only mods, reverse engineering and recon need no fal key. Images can also be made locally with ComfyUI via um comfy.
What does the AI agent cost?
That depends on the agent: Claude Code needs a Claude Pro, Max, Team or Enterprise plan or API billing; Codex needs a ChatGPT plan or API key; Gemini CLI has a free tier. Modding sessions are long, so on subscription plans expect to hit usage limits on bigger projects.
Will the agent spend my fal credit without asking?
The fal-assets skill tells the agent to check prices with um fal price and tell you the rough cost before about 20 generations or any video-sized job. Set a spending limit in your fal dashboard as a backstop.
Are these prices current?
They were checked against fal's model pages on October 9, 2026. fal changes prices and default models move, so run um fal price <endpoint> for the live number.