Add cross-assembly plugin system, texture optimizer, and multi-backend local AI pipeline

[Copilot]

Implements three major extensions to TextureCocktail: a proper open plugin API, a texture import settings analyzer/fixer, and a multi-backend local AI integration.

Plugin System

The previous LoadShaderWindow used Type.GetType("LuticaLab.TextureCocktail." + name) — only finding types in the hardcoded namespace. Third-party plugins from external assemblies were silently ignored.

New: TextureCocktailPluginRegistry scans all AppDomain assemblies on [InitializeOnLoadMethod], matching TextureCocktailContent subclasses by class name against the shader's last path segment. No registration step required.

// Any assembly, any namespace — auto-discovered
[TextureCocktailPlugin("Vignette", "Adds vignette darkening", "YourName", "1.0.0")]
public class VignetteEffect : TextureCocktailContent
{
    public override bool UseDefaultLayout => false;
    public override void OnGUI() { /* custom IMGUI */ }
    public override void OnShaderValueChanged() { }
}

• TextureCocktailPluginAttribute — optional metadata (display name, description, author, version)
• PluginBrowserWindow — LuticaLab → TextureCocktail Plugin Browser lists all discovered plugins
• PLUGIN_GUIDE.md — developer reference for plugin authoring

Texture Optimizer

New editor window (LuticaLab → Texture Optimizer) that batch-analyzes Texture2D assets via TextureImporter:

• Detects: non-POT dimensions, missing mipmaps on 3D textures, uncompressed formats, oversized textures, large compressed textures without crunch
• Platform profiles: Desktop / Mobile / VR (affects recommended compression format)
• One-click per-texture or batch "apply recommended fixes" (reimports assets)

Multi-Backend Local AI Connector

New editor window (LuticaLab → AI Connector) providing a text → text+image AI pipeline supporting multiple local AI backends:

Supported Backends

• Ollama (POST /api/generate, GET /api/tags) — supports vision models like llava
• OpenAI-Compatible APIs (POST /v1/chat/completions, GET /v1/models) — covers LocalAI, LM Studio, Jan, Kobold.cpp, llama.cpp server, text-generation-webui, and any other OpenAI-compatible endpoint

Architecture (Editor/AI/)

• AiBackendBase — abstract base class with AiRequest/AiResponse structs; add new backends by subclassing from any assembly
• AiTextureUtils — shared TextureToBase64 / Base64ToTexture utilities reusable across all backends
• OllamaBackend — Ollama-specific implementation using Newtonsoft.Json
• OpenAiCompatibleBackend — OpenAI-compatible implementation using Newtonsoft.Json

Features

• Backend selector dropdown; server URL pre-filled per backend
• Text prompt + optional Texture2D attachment (base64 PNG) for vision models
• Decodes and displays base64 images in response; allows saving to project assets
• Server URL and model persisted via EditorPrefs
• CancellationToken support

i18n

Added keys for all three features to English, Korean, and Japanese language files.

[LuticaCANARD]

Ollama외 다른 ai도 되게하라

[LuticaCANARD]

1. 전반적으로 로컬모델의 호환규격을 맞추려 한 부분은 좋으나, 재사용성을 감안하여 함수의 정의위치를 바꿔줄 것을 요청한다.

[LuticaCANARD]

이거 그냥 JSON Parser 쓰면 되지않나?

[Copilot]

Done in commit 984d4a8. The manual string-walking parsers have been replaced with Newtonsoft.Json (JObject.Parse / JArray) in both OllamaBackend and OpenAiCompatibleBackend — the same library already used by LanguageDisplayer.cs.

[LuticaCANARD]

이 인코딩 과정 다른 함수에 없었는가? 없었다면 공통 도구함수로 이전시키고 다른 class에서도 쓸 수 있게하라.

[Copilot]

Moved in commit 984d4a8. TextureToBase64 and Base64ToTexture now live in Editor/AI/AiTextureUtils.cs as public static methods, usable from any class. Both OllamaBackend and OpenAiCompatibleBackend call them from there.

[Copilot]

1. 전반적으로 로컬모델의 호환규격을 맞추려 한 부분은 좋으나, 재사용성을 감안하여 함수의 정의위치를 바꿔줄 것을 요청한다.

Addressed in commit 984d4a8. AI-related logic has been split across purpose-specific files in Editor/AI/:

• AiTextureUtils.cs — shared texture encode/decode utilities
• AiBackendBase.cs — abstract base + AiRequest/AiResponse structs
• OllamaBackend.cs — Ollama-specific HTTP logic
• OpenAiCompatibleBackend.cs — OpenAI-compatible logic

OllamaConnector.cs is now a thin UI window that delegates all network work to the chosen backend.

Ollama외 다른 ai도 되게하라

Added in commit 984d4a8. The window (LuticaLab → AI Connector) now has a Backend dropdown. Currently supported:

• Ollama — POST /api/generate (default: localhost:11434)
• OpenAI-Compatible — POST /v1/chat/completions (covers LocalAI, LM Studio, Jan, Kobold.cpp, llama.cpp server, text-generation-webui, etc.)

Additional backends can be added by creating a new class that inherits AiBackendBase in any assembly — the dropdown picks it up automatically.