Appearance
Architecture reading map
Start at the CLI surface, and then follow one runtime family at a time. This map gives contributors a reliable reading order without explaining every model detail.
For the broader documentation set, start at the mere.run documentation home.
Start here
- CLI entry point:
Sources/MereRunCLI/MereRunCLI.swift - Shared CLI helpers:
Sources/MereRunCLI/Support - Model paths and manifests:
Sources/MereRunCore/MereRunModelPaths.swiftSources/MereRunCore/MereRunModelManifest.swiftSources/MereRunCore/ModelResolver.swift
If you want to understand what a command does end to end, start at the command file, and then use the table to open the family entry point.
Image families
Klein image generation:
- CLI:
Sources/MereRunCLI/Commands/ImageGenerateCommand.swift - Runtime entry point:
Sources/MereRunCore/Flux2Klein/Flux2KleinGenerator.swift - Read next:
Sources/MereRunCore/Flux2Klein/Flux2KleinGenerator+ModelLoading.swiftSources/MereRunCore/Flux2Klein/Flux2KleinGenerator+Generation.swiftSources/MereRunCore/Flux2Klein/Flux2KleinGenerator+Chat.swift
ZImage generation:
- Runtime entry point:
Sources/MereRunCore/ZImageTurbo/ZImageTurboGenerator.swift - Read next:
Sources/MereRunCore/ZImageTurbo/ZImageTurboGenerator+ModelLoading.swiftSources/MereRunCore/ZImageTurbo/ZImageTurboGenerator+Inference.swiftSources/MereRunCore/ZImageTurbo/ZImageTurboGenerator+LoRA.swift
Qwen image editing:
- Runtime entry point:
Sources/MereRunCore/QwenImageEdit/QwenImageEditGenerator.swift - Read next:
Sources/MereRunCore/QwenImageEdit/QwenImageEditGenerator+ModelLoading.swiftSources/MereRunCore/QwenImageEdit/QwenImageEditGenerator+Encoding.swift
Krea 2 generation and LoRA training:
- Runtime entry point:
Sources/MereRunCore/Krea2/Krea2Generator.swift - Read next:
Sources/MereRunCore/Krea2/Krea2RawResources.swiftSources/MereRunCore/Krea2/Krea2Resources.swiftSources/MereRunCore/Krea2/Krea2LoRAInjector.swiftSources/MereRunCore/Krea2/Krea2LoRATrainer.swiftSources/MereRunCore/Krea2/Krea2Configs.swiftSources/MereRunCore/Krea2/Krea2Model.swiftSources/MereRunCore/Krea2/Krea2ModelLoader.swiftSources/MereRunCore/Krea2/Krea2SampleBuilder.swift
Shared text encoder stack used by image models:
- Public entry point:
Sources/MereRunCore/ZImageTurbo/Model/TextEncoder/TextEncoder.swift - Architecture internals:
Sources/MereRunCore/ZImageTurbo/Model/TextEncoder/TextEncoder+RoPE.swiftSources/MereRunCore/ZImageTurbo/Model/TextEncoder/TextEncoder+Blocks.swift
Speech stack
Speech synthesis command path:
- CLI:
Sources/MereRunCLI/Commands/SpeechSynthesizeCommand.swift - Runtime entry point:
Sources/AudioTTS/Qwen3TTS/Qwen3TTSGenerator.swift - Read next:
Sources/AudioTTS/Qwen3TTS/Qwen3TTSGenerator+Loading.swiftSources/AudioTTS/Qwen3TTS/Qwen3TTSGenerator+Generation.swiftSources/AudioTTS/Qwen3TTS/Qwen3TTSGenerator+Support.swift
Speech tokenizer internals:
- Public tokenizer surface:
Sources/AudioTTS/Qwen3TTS/Qwen3TTSSpeechTokenizer.swift - Decoder stack:
Sources/AudioTTS/Qwen3TTS/Qwen3TTSSpeechTokenizer+Decoder.swift - Encoder stack:
Sources/AudioTTS/Qwen3TTS/Qwen3TTSSpeechTokenizer+Encoder.swift
Speech transcription:
- CLI:
Sources/MereRunCLI/Commands/SpeechTranscribeCommand.swift - Runtime roots:
Sources/AudioSTT/Qwen3ASR/Qwen3ASRGenerator.swiftSources/AudioSTT/Parakeet/ParakeetGenerator.swift
Speaker diarization:
- CLI:
Sources/MereRunCLI/Commands/SpeechDiarizeCommand.swift - Runtime root:
Sources/AudioSTT/Sortformer/SortformerDiarizer.swift - Model and feature stack:
Sources/AudioSTT/Sortformer/SortformerModel.swift
OCR and vision
OCR:
- CLI:
Sources/MereRunCLI/Commands/VisionOCRCommand.swift - Runtime entry point:
Sources/MereRunCore/LightOnOCR/LightOnOCRGenerator.swift - Read next:
Sources/MereRunCore/LightOnOCR/LightOnOCRGenerator+Loading.swiftSources/MereRunCore/LightOnOCR/LightOnOCRGenerator+Inference.swiftSources/MereRunCore/LightOnOCR/LightOnOCRSupport.swift
Captioning and inspect flows:
- CLI:
Sources/MereRunCLI/Commands/VisionCaptionCommand.swiftSources/MereRunCLI/Commands/VisionInspectCommand.swiftSources/MereRunCLI/Commands/VisionSegmentCommand.swift
- Runtime roots:
Sources/MereRunCore/VLM/QwenVLCaptioner.swiftSources/MereRunCore/VLM/Qwen3VLAutoCaptioner.swiftSources/MereRunCore/VLM/QwenVLEncoder.swiftSources/MereRunCore/ZImageTurbo/Model/TextEncoder/Vision/QwenVisionAttention.swift
Segmentation and tracking runtime:
- CLI:
Sources/MereRunCLI/Commands/VisionSegmentCommand.swiftSources/MereRunCLI/Commands/VisionTrackCommand.swiftSources/MereRunCLI/Commands/VisionTrackLiveCommand.swift
- Native runtime:
Sources/MereRunCore/SAM3/SAM31Config.swiftSources/MereRunCore/SAM3/SAM31Resources.swiftSources/MereRunCore/SAM3/SAM31Tokenizer.swiftSources/MereRunCore/SAM3/SAM31Model.swiftSources/MereRunCore/SAM3/SAM31InteractiveSAM.swiftSources/MereRunCore/SAM3/SAM31Prompts.swiftSources/MereRunCore/SAM3/SAM31ImageSegmenter.swiftSources/MereRunCore/SAM3/SAM31VideoIO.swiftSources/MereRunCore/SAM3/SAM31VideoTracker.swiftSources/MereRunCore/SAM3/SAM31CameraCapture.swift
Native object reconstruction:
- CLI:
Sources/MereRunCLI/Commands/VisionImageTo3DCommand.swiftSources/MereRunCLI/Commands/VisionImageTo3DTrellis2Command.swiftSources/MereRunCLI/Commands/ImageReconstruct3DMultiviewCommand.swift
- Shared mesh contract:
Sources/MereRunCore/Asset3D - Native runtimes:
Sources/MereRunCore/TripoSR/TripoSRGenerator.swiftSources/MereRunCore/Trellis2/Trellis2Generator.swiftSources/MereRunCore/InstantMesh/InstantMeshGenerator.swift
Music, SFX, and video
Music generation:
- CLI:
Sources/MereRunCLI/Commands/MusicGenerateCommand.swift - Runtime entry point:
Sources/MereRunCore/ACEStep/ACEStepPipeline.swift - Read next:
Sources/MereRunCore/ACEStep/ACEStepPipeline+Prompting.swiftSources/MereRunCore/ACEStep/ACEStepPipeline+Generation.swift
Music source separation:
- CLI:
Sources/MereRunCLI/Commands/MusicSeparateCommand.swift - Runtime entry point:
Sources/MereRunCore/RoFormer/RoFormerSeparator.swift - Read next:
Sources/MereRunCore/RoFormer/RoFormerResources.swiftSources/MereRunCore/RoFormer/BSRoFormer.swiftSources/MereRunCore/RoFormer/RoFormerDSP.swift
Audio bandwidth extension and super-resolution:
- CLI:
Sources/MereRunCLI/Commands/AudioEnhanceCommand.swift - Runtime entry point:
Sources/MereRunCore/APBWE/APBWEEnhancer.swift - General-audio entry point:
Sources/MereRunCore/UniverSR/UniverSREnhancer.swift - Read next:
Sources/MereRunCore/APBWE/APBWEResources.swiftSources/MereRunCore/APBWE/APBWEModel.swiftSources/MereRunCore/UniverSR/UniverSRResources.swiftSources/MereRunCore/UniverSR/UniverSRModel.swiftdocs/architecture/audio-enhancement-ap-bwe-report.mddocs/architecture/audio-enhancement-universr-report.md
Sound-effect generation:
- CLI:
Sources/MereRunCLI/Commands/SFXGenerateCommand.swift - Runtime root:
Sources/MereRunCore/Woosh/WooshGenerator.swift - Read next:
Sources/MereRunCore/Woosh/WooshDiT.swiftSources/MereRunCore/Woosh/WooshRobertaTextEncoder.swiftSources/MereRunCore/Woosh/WooshVocosDecoder.swift
Video generation:
- CLI:
Sources/MereRunCLI/Commands/VideoCommand.swift(VideoGenerateandVideoExportLatents)
- Runtime root:
Sources/MereRunCore/LTX/LTXDistilledLatentGenerator.swift - Read next:
Sources/MereRunCore/LTX/LTXGemmaTextEncoder.swiftSources/MereRunCore/LTX/LTXVideoMP4Writer.swift
The LTX runtime combines public generation flow with lower-level model definitions. Read it in this order:
- public generation types and
LTXDistilledLatentGenerator - diffusion helpers such as
denoiseLoop, latent conditioning, and image loading - decoder and encoder support modules
- unified AV generator types near the end
Cosmos3-Edge omnimodal generation and world simulation:
- CLI generation and reasoner:
Sources/MereRunCLI/Commands/VideoCosmos3Command.swift - Persistent world server:
Sources/MereRunCLI/Commands/WorldCommand.swift - Runtime entry points:
Sources/MereRunCore/Cosmos3/Cosmos3EdgeGenerator.swiftSources/MereRunCore/Cosmos3/Cosmos3Reasoner.swiftSources/MereRunCore/Cosmos3/Cosmos3WorldSession.swift
- Read next:
Sources/MereRunCore/Cosmos3/Cosmos3Resources.swiftSources/MereRunCore/Cosmos3/Cosmos3Action.swiftSources/MereRunCore/Cosmos3/Cosmos3Sequence.swiftSources/MereRunCore/Cosmos3/Cosmos3Transformer.swiftSources/MereRunCore/Cosmos3/Cosmos3ModelLoader.swiftSources/MereRunCore/Cosmos3/Cosmos3Scheduler.swiftSources/MereRunCore/Cosmos3/Cosmos3ReasonerVision.swift
Validation and diagnostics
The advanced image validator is intentionally separate from generation:
- CLI root:
Sources/MereRunCLI/Commands/ImageValidateCommand.swift - Read next:
Sources/MereRunCLI/Commands/ImageValidateCommand+VAE.swiftSources/MereRunCLI/Commands/ImageValidateCommand+EncoderTransformer.swiftSources/MereRunCLI/Commands/ImageValidateCommand+Pipeline.swift
Contributor reading order
- Start with
Sources/MereRunCLI/MereRunCLI.swift. - Pick one modality and read the matching command file.
- Jump to that modality's runtime entry point.
- Follow companion files in order: loading or preparation, then generation, then output.
- If you need implementation detail, continue to the larger model-definition files.