A desktop editor for Diablo IV loot filter share codes — import, edit visually, export back to the game.
D4's in-game filter UI is functional but tedious for anything beyond the simplest rules. D4LootBench lets you import a share code, edit all its rules with a proper visual editor, and export a new share code to paste back into the game.
Single .exe, no installer, Windows only. Copy it anywhere and run it.
First run: Windows SmartScreen may warn that the file is from an unknown publisher. This is expected for unsigned community tools — click More info → Run anyway to proceed.
- Import any D4 loot filter share code
- Generate a BiS filter from a build guide — paste a gear section from Mobalytics, Maxroll, or Icy Veins and get a working filter in seconds
- Add, remove, reorder, and edit rules with a full visual editor
- Edit all condition types: rarity, item power, affixes, greater affixes, codex of power, item types, unique items, talisman sets, and item properties
- Color-code rules with a full HSV color picker
- Export back to a share code to paste into D4
- Raw JSON editor for bulk edits and power users
- Optional AI assistant (Ollama) — describe a rule in plain English, get a filter rule back
- Open D4's in-game Loot Filter menu → select your filter → Share → copy the share code
- Paste the code into D4LootBench
- Edit rules and conditions in the visual editor (or switch to the Raw JSON editor)
- Click Copy Code → paste back into D4's share code field
- Open a build guide on Mobalytics, Maxroll, or Icy Veins and navigate to the gear / Best in Slot section
- Select and copy the gear table (Ctrl+C)
- Click Import from Build Guide in the D4LootBench toolbar
- Paste the text, confirm the detected format, and click Import Filter
- Tune the generated rules in the visual editor — by default each slot rule requires 2 of the 4 top affixes
The AI assistant is an opt-in feature — D4LootBench works fully without it.
Recommended setup: Ollama (free, runs locally)
- Install Ollama from ollama.com
- Pull a model (see table below):
ollama pull qwen2.5-coder:14b - Open D4LootBench → click the AI button in the toolbar to expand the panel
- Set Base URL to
http://localhost:11434and Model to the name you pulled
Then describe what you want in plain English, e.g. "show all ancestral items with a greater affix", and the assistant will generate a filter rule you can review before adding.
Recommended models
| GPU VRAM | Recommended model | Notes |
|---|---|---|
| 10 GB+ | qwen2.5-coder:14b |
Best quality; 11/11 on structured rule generation |
| 6–9 GB | qwen2.5-coder:7b |
Good quality (11/11); fits most mid-range GPUs |
If you're unsure how much VRAM your GPU has, check Task Manager → Performance → GPU. If a model is too large to fit in VRAM, Ollama will silently fall back to CPU — inference will work but will be significantly slower.
Other models: Non-coding general-purpose models (e.g.
llama3.2) are not recommended. The assistant output is structured JSON that must match a strict schema — general models tested poorly and produced incorrect rules even with precise prompts. Use a coder model if at all possible. If you choose to experiment with other models, review every generated rule carefully before adding it to your filter.
Cloud providers (Anthropic, OpenAI) are not yet wired into the UI. The provider abstraction is in place and may be added based on community interest.
D4LootBench embeds a copy of d4-data.json — the database of affix names, item types, unique items, skills, and talisman sets used to populate the editor pickers.
To edit it (e.g. to add a newly released item or correct a name):
- File → Export d4-data.json — writes the embedded file next to
D4LootBench.exe - Edit the file with any text editor
- Restart D4LootBench — the local copy takes precedence over the embedded one
See docs/d4-data-format.md for the full schema reference. Community corrections are welcome — see CONTRIBUTING.md.
Filter code won't import / produces unexpected results Re-export the filter from D4's in-game UI instead of using a code shared externally. Codes shared via community sites or older tools may have subtle encoding differences from the current game version.
AI assistant not responding
Verify Ollama is running (ollama list in a terminal). Confirm the Base URL in the AI panel matches your Ollama port — the default is http://localhost:11434.
Requires .NET 10 SDK.
dotnet build # build the full solution
dotnet test # run the test suite (88 tests)
# produce a self-contained single-file exe
dotnet publish src/D4LootBench.App -r win-x64 -p:PublishSingleFile=true --self-contained trueFor those interested in the implementation:
- Custom protobuf codec (~80 lines, 3 wire types) — reverse-engineered from D4's binary share code format; no Google.Protobuf dependency, handles unknown fields gracefully for patch resilience
- Clean three-library solution —
D4LootBench.Core(zero WPF dependency),D4LootBench.Ai(zero WPF dependency),D4LootBench.App(WPF shell) - MVVM with CommunityToolkit.Mvvm source generators and Microsoft.Extensions.DependencyInjection
ILlmProviderabstraction over Ollama with clean extension points for additional providers- Annotated
{id, name}JSON format — human-readable and LLM-interpretable while keeping hash IDs (SNO IDs) authoritative - 88 unit tests covering codec round-trips, validation rules, annotated JSON serialization, and build guide parser coverage across all three formats
See docs/filter-format.md for the full protocol buffer format specification.
D4LootBench was built on the shoulders of the following community reverse-engineering work:
| Project | License | Contribution |
|---|---|---|
| Upsilon72/d4-filter-generator | MIT | Protobuf wire format, condition type encoding, affix hash IDs |
| fnuecke/diablo4-loot-filter-viewer | Unlicense | Complete .proto field layout, all 10 condition type semantics, names.json ID tables |
| DiabloTools/d4data | MIT | CoreTOC_flat.json — authoritative datamined SNO IDs for skills, item types, affixes, and unique items |
| d4lfteam/d4lf | MIT | Affix name reference database |
| Raxx (Raxxanterax) | — | Real-world filter export used to validate and extend the format specification |
MIT — see LICENSE.
D4LootBench is an unofficial community tool and is not affiliated with Blizzard Entertainment.