Engine interface

Each engine exports key, label, isReady(), and generate({ characterPhotoPaths, posePhotoPath, prompt, outputPath, outputSettings, apiKey, model }). isReady() returns { ready, reason? }; generate must write an image to outputPath or throw a clear Error. outputSettings and model are optional and adapters must provide safe defaults.

characterPhotoPaths is an array of 1-4 local file paths (one per person in the generation, in order — this is also the order the prompt text refers to them by, e.g. "Image 1", "Image 2"). Every path must be included in the request to the underlying model; dropping any of them means that person won't appear in the output. prompt (built by lib/promptTemplate.js) already accounts for however many paths were given, so adapters don't need to reason about the count themselves — just forward every image, in order, followed by the pose reference.

Advanced Studio

An engine appears in Advanced Studio when it declares what it can do there:

• Images: capabilities.freeform, a generateFreeform({ referencePaths, prompt, outputPath, outputSettings, apiKey, model }) method, and imageSettings: { aspectRatios, resolutions: [{ id, label, quality }], defaultResolution, note? }. The node menus are built from imageSettings; quality is the value outputSettings.quality receives.
• Video: capabilities.video, generateVideo(...), videoModels and modelDefinition(modelId).
• Text (LLM Text nodes): assistPrompt({ instruction, system }) returning { text }. system replaces the default prompt-assistant instructions when given.
• Image edits (Image Transform nodes): transformImage({ inputPath, instruction, context, outputPath }) writes a PNG to outputPath after editing the image at inputPath as instruction says, using context (connected text) as material. outputPath is in a scratch folder the caller removes.

Register the module in engines/index.js; nothing else needs to change.