commit 263cb060eeaf8782aeaf6c33ffb67f042d34f70a
Author: Pharma-Lobby <happyhutmacher3@gmail.com>
Date:   Fri Jul 10 16:06:22 2026 +0000

    ComfyUI-Replicate-Select v0.1.0 — 14 kuratierte Modelle + ReplicateCustom
    
    Bild (Seedream 4.5, Nano Banana Pro), Video (Kling v3, Veo 3.1, Seedance 2.0 Fast),
    Upscale (SUPIR, Real-ESRGAN), TTS (MiniMax 2.8 HD, Inworld TTS 2, ElevenLabs v3),
    Lipsync (VEED Fabric, Pruna Avatar, Sync 2 Pro, HeyGen Precision).
    
    Architektur nach Runware-Vorbild: nodes.json + Generator + Escape-Hatch.
    Katalog-Refresh ohne Token via scripts/build_catalog.py (public Schema-Seiten).
    
    Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
    Claude-Session: https://claude.ai/code/session_01K33rz4XBNkq2gKoQov5D37

diff --git a/README.md b/README.md
new file mode 100644
index 0000000..1966835
--- /dev/null
+++ b/README.md
@@ -0,0 +1,43 @@
+# ComfyUI-Replicate-Select
+
+Kuratierte Replicate-Modelle als ComfyUI-Nodes — schlank statt Katalog-Flut.
+Architektur nach dem Vorbild von [ComfyUI-Runware](https://github.com/Runware/ComfyUI-Runware):
+`nodes.json` (Katalog) + Generator + `ReplicateCustom` als Escape-Hatch.
+
+## Modelle (14)
+
+| Kategorie | Nodes |
+|---|---|
+| image | Seedream 4.5 · Nano Banana Pro |
+| video | Kling v3 Video · Veo 3.1 · Seedance 2.0 Fast |
+| upscale | SUPIR · Real-ESRGAN |
+| tts | MiniMax Speech 2.8 HD · Inworld TTS 2 · ElevenLabs v3 |
+| lipsync | VEED Fabric 1.0 · Pruna Video Avatar · Sync Lipsync 2 Pro · HeyGen Lipsync Precision |
+
+Dazu: **Replicate Custom** — beliebige Modell-ID + JSON-Input.
+
+## Auth
+
+`REPLICATE_API_TOKEN` als Umgebungsvariable (auf Modal: Secret `mcp-keys`).
+Kein Key im Node, keine Key-Dateien.
+
+## Outputs
+
+- **image** → IMAGE-Tensor (direkt weiterverdrahtbar)
+- **audio** → AUDIO (ComfyUI-nativ, z.B. → SaveAudio oder → Lipsync-Nodes via Upload)
+- **video** → MP4 im Output-Ordner, Node gibt den Pfad als STRING zurück
+
+## Katalog aktualisieren / Modelle hinzufügen
+
+```bash
+# Modell in scripts/build_catalog.py → META eintragen, dann:
+python scripts/build_catalog.py
+```
+
+Zieht die Input-Schemas von den öffentlichen Replicate-Schema-Seiten — kein Token nötig.
+
+## Hinweise
+
+- Bild-Inputs werden als PNG hochgeladen (klein → data-URI, groß → Replicate Files-API)
+- Audio-Inputs (Lipsync) akzeptieren ComfyUI-AUDIO und werden als WAV hochgeladen
+- Video-Inputs erwarten eine URL (STRING) — Replicate braucht abrufbare Dateien
diff --git a/__init__.py b/__init__.py
new file mode 100644
index 0000000..39a8c6b
--- /dev/null
+++ b/__init__.py
@@ -0,0 +1,3 @@
+from .nodes import NODE_CLASS_MAPPINGS, NODE_DISPLAY_NAME_MAPPINGS
+
+__all__ = ["NODE_CLASS_MAPPINGS", "NODE_DISPLAY_NAME_MAPPINGS"]
diff --git a/__pycache__/nodes.cpython-311.pyc b/__pycache__/nodes.cpython-311.pyc
new file mode 100644
index 0000000..d957b89
Binary files /dev/null and b/__pycache__/nodes.cpython-311.pyc differ
diff --git a/nodes.json b/nodes.json
new file mode 100644
index 0000000..cca7a96
--- /dev/null
+++ b/nodes.json
@@ -0,0 +1,1486 @@
+{
+ "models": [
+  {
+   "id": "bytedance/seedream-4.5",
+   "name": "Seedream 4.5",
+   "category": "image",
+   "output": "image",
+   "inputs": [
+    {
+     "name": "prompt",
+     "order": 0,
+     "required": true,
+     "description": "Text prompt for image generation. Maximum 4000 characters. BytePlus recommends keeping prompts under 600 English words for best results.",
+     "kind": "string",
+     "default": "",
+     "multiline": true
+    },
+    {
+     "name": "image_input",
+     "order": 1,
+     "required": false,
+     "description": "Input image(s) for image-to-image generation. List of 1-14 images for single or multi-reference generation.",
+     "kind": "image",
+     "is_list": true
+    },
+    {
+     "name": "size",
+     "order": 2,
+     "required": false,
+     "description": "Image resolution: 2K (2048px), 4K (4096px), or 'custom' for specific dimensions. Note: 1K resolution is not supported in Seedream 4.5.",
+     "kind": "combo",
+     "options": [
+      "2K",
+      "4K",
+      "custom"
+     ],
+     "default": "2K"
+    },
+    {
+     "name": "aspect_ratio",
+     "order": 3,
+     "required": false,
+     "description": "Image aspect ratio. Only used when size is not 'custom'. Use 'match_input_image' to automatically match the input image's aspect ratio.",
+     "kind": "combo",
+     "options": [
+      "match_input_image",
+      "1:1",
+      "4:3",
+      "3:4",
+      "4:5",
+      "5:4",
+      "16:9",
+      "9:16",
+      "3:2",
+      "2:3",
+      "21:9",
+      "9:21"
+     ],
+     "default": "match_input_image"
+    },
+    {
+     "name": "width",
+     "order": 4,
+     "required": false,
+     "description": "Custom image width (only used when size='custom'). Range: 1024-4096 pixels.",
+     "kind": "int",
+     "default": 2048,
+     "min": 1024,
+     "max": 4096
+    },
+    {
+     "name": "height",
+     "order": 5,
+     "required": false,
+     "description": "Custom image height (only used when size='custom'). Range: 1024-4096 pixels.",
+     "kind": "int",
+     "default": 2048,
+     "min": 1024,
+     "max": 4096
+    },
+    {
+     "name": "sequential_image_generation",
+     "order": 6,
+     "required": false,
+     "description": "Group image generation mode. 'disabled' generates a single image. 'auto' lets the model decide whether to generate multiple related images (e.g., story scenes, character variations).",
+     "kind": "combo",
+     "options": [
+      "disabled",
+      "auto"
+     ],
+     "default": "disabled"
+    },
+    {
+     "name": "max_images",
+     "order": 7,
+     "required": false,
+     "description": "Maximum number of images to generate when sequential_image_generation='auto'. Range: 1-15. Total images (input + generated) cannot exceed 15.",
+     "kind": "int",
+     "default": 1,
+     "min": 1,
+     "max": 15
+    },
+    {
+     "name": "disable_safety_checker",
+     "order": 8,
+     "required": false,
+     "description": "Disable the safety checker for generated images. When enabled, input moderation is relaxed to only block illegal content (CSAM). Use responsibly.",
+     "kind": "bool",
+     "default": false
+    }
+   ]
+  },
+  {
+   "id": "google/nano-banana-pro",
+   "name": "Nano Banana Pro",
+   "category": "image",
+   "output": "image",
+   "inputs": [
+    {
+     "name": "prompt",
+     "order": 0,
+     "required": true,
+     "description": "A text description of the image you want to generate",
+     "kind": "string",
+     "default": "",
+     "multiline": true
+    },
+    {
+     "name": "image_input",
+     "order": 1,
+     "required": false,
+     "description": "Input images to transform or use as reference (supports up to 14 images)",
+     "kind": "image",
+     "is_list": true
+    },
+    {
+     "name": "aspect_ratio",
+     "order": 2,
+     "required": false,
+     "description": "Aspect ratio of the generated image",
+     "kind": "combo",
+     "options": [
+      "match_input_image",
+      "1:1",
+      "2:3",
+      "3:2",
+      "3:4",
+      "4:3",
+      "4:5",
+      "5:4",
+      "9:16",
+      "16:9",
+      "21:9"
+     ],
+     "default": "match_input_image"
+    },
+    {
+     "name": "resolution",
+     "order": 3,
+     "required": false,
+     "description": "Resolution of the generated image",
+     "kind": "combo",
+     "options": [
+      "1K",
+      "2K",
+      "4K"
+     ],
+     "default": "2K"
+    },
+    {
+     "name": "output_format",
+     "order": 4,
+     "required": false,
+     "description": "Format of the output image",
+     "kind": "combo",
+     "options": [
+      "jpg",
+      "png"
+     ],
+     "default": "jpg"
+    },
+    {
+     "name": "safety_filter_level",
+     "order": 5,
+     "required": false,
+     "description": "block_low_and_above is strictest, block_medium_and_above blocks some prompts, block_only_high is most permissive but some prompts will still be blocked",
+     "kind": "combo",
+     "options": [
+      "block_low_and_above",
+      "block_medium_and_above",
+      "block_only_high"
+     ],
+     "default": "block_only_high"
+    },
+    {
+     "name": "allow_fallback_model",
+     "order": 6,
+     "required": false,
+     "description": "Fallback to another model (currently bytedance/seedream-5) if Nano Banana Pro is at capacity.",
+     "kind": "bool",
+     "default": false
+    }
+   ]
+  },
+  {
+   "id": "kwaivgi/kling-v3-video",
+   "name": "Kling v3 Video",
+   "category": "video",
+   "output": "video",
+   "inputs": [
+    {
+     "name": "prompt",
+     "order": 0,
+     "required": true,
+     "description": "Text prompt for video generation. Max 2500 characters.",
+     "kind": "string",
+     "default": "",
+     "multiline": true
+    },
+    {
+     "name": "negative_prompt",
+     "order": 1,
+     "required": false,
+     "description": "Things you do not want to see in the video. Max 2500 characters.",
+     "kind": "string",
+     "default": "",
+     "multiline": true
+    },
+    {
+     "name": "start_image",
+     "order": 2,
+     "required": false,
+     "description": "First frame image. Supports .jpg/.jpeg/.png, max 10MB, min 300px, aspect ratio 1:2.5 to 2.5:1.",
+     "kind": "image",
+     "is_list": false
+    },
+    {
+     "name": "end_image",
+     "order": 3,
+     "required": false,
+     "description": "Last frame image. Requires start_image. Supports .jpg/.jpeg/.png, max 10MB, min 300px.",
+     "kind": "image",
+     "is_list": false
+    },
+    {
+     "name": "mode",
+     "order": 4,
+     "required": false,
+     "description": "'standard' generates 720p, 'pro' generates 1080p, '4k' generates 4K.",
+     "kind": "combo",
+     "options": [
+      "standard",
+      "pro",
+      "4k"
+     ],
+     "default": "pro"
+    },
+    {
+     "name": "aspect_ratio",
+     "order": 5,
+     "required": false,
+     "description": "Aspect ratio. Ignored when start_image is provided.",
+     "kind": "combo",
+     "options": [
+      "16:9",
+      "9:16",
+      "1:1"
+     ],
+     "default": "16:9"
+    },
+    {
+     "name": "duration",
+     "order": 6,
+     "required": false,
+     "description": "Video duration in seconds.",
+     "kind": "int",
+     "default": 5,
+     "min": 3,
+     "max": 15
+    },
+    {
+     "name": "generate_audio",
+     "order": 7,
+     "required": false,
+     "description": "Generate native audio for the video.",
+     "kind": "bool",
+     "default": false
+    },
+    {
+     "name": "multi_prompt",
+     "order": 8,
+     "required": false,
+     "description": "JSON array of shot definitions for multi-shot mode. Each shot: {\"prompt\": \"...\", \"duration\": N}. Max 6 shots, min 1s per shot, total must equal duration.",
+     "kind": "string",
+     "default": "",
+     "multiline": true
+    }
+   ]
+  },
+  {
+   "id": "google/veo-3.1",
+   "name": "Veo 3.1",
+   "category": "video",
+   "output": "video",
+   "inputs": [
+    {
+     "name": "prompt",
+     "order": 0,
+     "required": true,
+     "description": "Text prompt for video generation",
+     "kind": "string",
+     "default": "",
+     "multiline": true
+    },
+    {
+     "name": "aspect_ratio",
+     "order": 1,
+     "required": false,
+     "description": "Video aspect ratio",
+     "kind": "combo",
+     "options": [
+      "16:9",
+      "9:16"
+     ],
+     "default": "16:9"
+    },
+    {
+     "name": "duration",
+     "order": 2,
+     "required": false,
+     "description": "Video duration in seconds",
+     "kind": "combo",
+     "options": [
+      4,
+      6,
+      8
+     ],
+     "default": 8
+    },
+    {
+     "name": "image",
+     "order": 3,
+     "required": false,
+     "description": "Input image to start generating from. Ideal images are 16:9 or 9:16 and 1280x720 or 720x1280, depending on the aspect ratio you choose.",
+     "kind": "image",
+     "is_list": false
+    },
+    {
+     "name": "last_frame",
+     "order": 4,
+     "required": false,
+     "description": "Ending image for interpolation. When provided with an input image, creates a transition between the two images.",
+     "kind": "image",
+     "is_list": false
+    },
+    {
+     "name": "reference_images",
+     "order": 5,
+     "required": false,
+     "description": "1 to 3 reference images for subject-consistent generation (reference-to-video, or R2V). Reference images only work with 16:9 aspect ratio and 8-second duration. Last frame is ignored if reference imag",
+     "kind": "image",
+     "is_list": true
+    },
+    {
+     "name": "negative_prompt",
+     "order": 6,
+     "required": false,
+     "description": "Description of what to exclude from the generated video",
+     "kind": "string",
+     "default": "",
+     "multiline": true
+    },
+    {
+     "name": "resolution",
+     "order": 7,
+     "required": false,
+     "description": "Resolution of the generated video",
+     "kind": "combo",
+     "options": [
+      "720p",
+      "1080p"
+     ],
+     "default": "1080p"
+    },
+    {
+     "name": "generate_audio",
+     "order": 8,
+     "required": false,
+     "description": "Generate audio with the video",
+     "kind": "bool",
+     "default": true
+    },
+    {
+     "name": "seed",
+     "order": 9,
+     "required": false,
+     "description": "Random seed. Omit for random generations",
+     "kind": "int",
+     "default": 0,
+     "min": -2147483648,
+     "max": 2147483647
+    }
+   ]
+  },
+  {
+   "id": "bytedance/seedance-2.0-fast",
+   "name": "Seedance 2.0 Fast",
+   "category": "video",
+   "output": "video",
+   "inputs": [
+    {
+     "name": "prompt",
+     "order": 0,
+     "required": true,
+     "description": "Text prompt for video generation",
+     "kind": "string",
+     "default": "",
+     "multiline": true
+    },
+    {
+     "name": "image",
+     "order": 1,
+     "required": false,
+     "description": "Input image for image-to-video generation (first frame). Cannot be combined with reference images.",
+     "kind": "image",
+     "is_list": false
+    },
+    {
+     "name": "last_frame_image",
+     "order": 2,
+     "required": false,
+     "description": "Input image for last frame generation. Only works if a first frame image is also provided. Cannot be combined with reference images.",
+     "kind": "image",
+     "is_list": false
+    },
+    {
+     "name": "reference_images",
+     "order": 3,
+     "required": false,
+     "description": "Reference images (up to 9) for character consistency, style guidance, and scene composition. Cannot be used together with first/last frame images. You can reference them in your prompt as [Image1], [I",
+     "kind": "image",
+     "is_list": true
+    },
+    {
+     "name": "reference_videos",
+     "order": 4,
+     "required": false,
+     "description": "Reference videos (up to 3, total duration max 15s) for motion transfer, style reference, and editing. Reference them in your prompt as [Video1], [Video2], etc.",
+     "kind": "uri_string"
+    },
+    {
+     "name": "reference_audios",
+     "order": 5,
+     "required": false,
+     "description": "Reference audio files (up to 3, total duration max 15s) for audio-driven generation and lip-sync. Requires at least one reference image or video. Reference them in your prompt as [Audio1], [Audio2], e",
+     "kind": "audio"
+    },
+    {
+     "name": "duration",
+     "order": 6,
+     "required": false,
+     "description": "Video duration in seconds. Set to -1 for intelligent duration (model picks the best length).",
+     "kind": "int",
+     "default": 5,
+     "min": -1,
+     "max": 15
+    },
+    {
+     "name": "resolution",
+     "order": 7,
+     "required": false,
+     "description": "Video resolution.",
+     "kind": "combo",
+     "options": [
+      "480p",
+      "720p"
+     ],
+     "default": "720p"
+    },
+    {
+     "name": "aspect_ratio",
+     "order": 8,
+     "required": false,
+     "description": "Video aspect ratio. Set to 'adaptive' to let the model choose the best ratio based on inputs.",
+     "kind": "combo",
+     "options": [
+      "16:9",
+      "4:3",
+      "1:1",
+      "3:4",
+      "9:16",
+      "21:9",
+      "9:21",
+      "adaptive"
+     ],
+     "default": "16:9"
+    },
+    {
+     "name": "generate_audio",
+     "order": 9,
+     "required": false,
+     "description": "Generate synchronized audio with the video, including dialogue (use double quotes in prompt), sound effects, and background music.",
+     "kind": "bool",
+     "default": true
+    },
+    {
+     "name": "seed",
+     "order": 10,
+     "required": false,
+     "description": "Random seed. Set for reproducible generation.",
+     "kind": "int",
+     "default": 0,
+     "min": -2147483648,
+     "max": 2147483647
+    }
+   ]
+  },
+  {
+   "id": "cjwbw/supir",
+   "name": "SUPIR Upscale",
+   "category": "upscale",
+   "output": "image",
+   "inputs": [
+    {
+     "name": "image",
+     "order": 1,
+     "required": true,
+     "description": "Low quality input image.",
+     "kind": "image",
+     "is_list": false
+    },
+    {
+     "name": "model_name",
+     "order": 0,
+     "required": false,
+     "description": "Choose a model. SUPIR-v0Q is the default training settings with paper. SUPIR-v0F is high generalization and high image quality in most cases. Training with light degradation settings. Stage1 encoder o",
+     "kind": "combo",
+     "options": [
+      "SUPIR-v0Q",
+      "SUPIR-v0F"
+     ],
+     "default": "SUPIR-v0Q"
+    },
+    {
+     "name": "upscale",
+     "order": 2,
+     "required": false,
+     "description": "Upsampling ratio of given inputs.",
+     "kind": "int",
+     "default": 1,
+     "min": -2147483648,
+     "max": 2147483647
+    },
+    {
+     "name": "min_size",
+     "order": 3,
+     "required": false,
+     "description": "Minimum resolution of output images.",
+     "kind": "float",
+     "default": 1024,
+     "min": -1000000000.0,
+     "max": 1000000000.0
+    },
+    {
+     "name": "edm_steps",
+     "order": 4,
+     "required": false,
+     "description": "Number of steps for EDM Sampling Schedule.",
+     "kind": "int",
+     "default": 50,
+     "min": 1,
+     "max": 500
+    },
+    {
+     "name": "use_llava",
+     "order": 5,
+     "required": false,
+     "description": "Use LLaVA model to get captions.",
+     "kind": "bool",
+     "default": true
+    },
+    {
+     "name": "a_prompt",
+     "order": 6,
+     "required": false,
+     "description": "Additive positive prompt for the inputs.",
+     "kind": "string",
+     "default": "Cinematic, High Contrast, highly detailed, taken using a Canon EOS R camera, hyper detailed photo - realistic maximum detail, 32k, Color Grading, ultra HD, extreme meticulous detailing, skin pore detailing, hyper sharpness, perfect without deformations.",
+     "multiline": true
+    },
+    {
+     "name": "n_prompt",
+     "order": 7,
+     "required": false,
+     "description": "Negative prompt for the inputs.",
+     "kind": "string",
+     "default": "painting, oil painting, illustration, drawing, art, sketch, oil painting, cartoon, CG Style, 3D render, unreal engine, blurring, dirty, messy, worst quality, low quality, frames, watermark, signature, jpeg artifacts, deformed, lowres, over-smooth",
+     "multiline": true
+    },
+    {
+     "name": "color_fix_type",
+     "order": 8,
+     "required": false,
+     "description": "Color Fixing Type..",
+     "kind": "combo",
+     "options": [
+      "None",
+      "AdaIn",
+      "Wavelet"
+     ],
+     "default": "Wavelet"
+    },
+    {
+     "name": "s_stage1",
+     "order": 9,
+     "required": false,
+     "description": "Control Strength of Stage1 (negative means invalid).",
+     "kind": "int",
+     "default": -1,
+     "min": -2147483648,
+     "max": 2147483647
+    },
+    {
+     "name": "s_churn",
+     "order": 10,
+     "required": false,
+     "description": "Original churn hy-param of EDM.",
+     "kind": "float",
+     "default": 5,
+     "min": -1000000000.0,
+     "max": 1000000000.0
+    },
+    {
+     "name": "s_noise",
+     "order": 11,
+     "required": false,
+     "description": "Original noise hy-param of EDM.",
+     "kind": "float",
+     "default": 1.003,
+     "min": -1000000000.0,
+     "max": 1000000000.0
+    },
+    {
+     "name": "s_cfg",
+     "order": 12,
+     "required": false,
+     "description": " Classifier-free guidance scale for prompts.",
+     "kind": "float",
+     "default": 7.5,
+     "min": 1,
+     "max": 20
+    },
+    {
+     "name": "s_stage2",
+     "order": 13,
+     "required": false,
+     "description": "Control Strength of Stage2.",
+     "kind": "float",
+     "default": 1,
+     "min": -1000000000.0,
+     "max": 1000000000.0
+    },
+    {
+     "name": "linear_CFG",
+     "order": 14,
+     "required": false,
+     "description": "Linearly (with sigma) increase CFG from 'spt_linear_CFG' to s_cfg.",
+     "kind": "bool",
+     "default": false
+    },
+    {
+     "name": "linear_s_stage2",
+     "order": 15,
+     "required": false,
+     "description": "Linearly (with sigma) increase s_stage2 from 'spt_linear_s_stage2' to s_stage2.",
+     "kind": "bool",
+     "default": false
+    },
+    {
+     "name": "spt_linear_CFG",
+     "order": 16,
+     "required": false,
+     "description": "Start point of linearly increasing CFG.",
+     "kind": "float",
+     "default": 1,
+     "min": -1000000000.0,
+     "max": 1000000000.0
+    },
+    {
+     "name": "spt_linear_s_stage2",
+     "order": 17,
+     "required": false,
+     "description": "Start point of linearly increasing s_stage2.",
+     "kind": "float",
+     "default": 0,
+     "min": -1000000000.0,
+     "max": 1000000000.0
+    },
+    {
+     "name": "seed",
+     "order": 18,
+     "required": false,
+     "description": "Random seed. Leave blank to randomize the seed",
+     "kind": "int",
+     "default": 0,
+     "min": -2147483648,
+     "max": 2147483647
+    }
+   ]
+  },
+  {
+   "id": "cjwbw/real-esrgan",
+   "name": "Real-ESRGAN Upscale",
+   "category": "upscale",
+   "output": "image",
+   "inputs": [
+    {
+     "name": "image",
+     "order": 0,
+     "required": true,
+     "description": "Input image",
+     "kind": "image",
+     "is_list": false
+    },
+    {
+     "name": "upscale",
+     "order": 1,
+     "required": false,
+     "description": "Upscaling factor",
+     "kind": "combo",
+     "options": [
+      2,
+      4,
+      8
+     ],
+     "default": 4
+    }
+   ]
+  },
+  {
+   "id": "minimax/speech-2.8-hd",
+   "name": "MiniMax Speech 2.8 HD",
+   "category": "tts",
+   "output": "audio",
+   "inputs": [
+    {
+     "name": "text",
+     "order": 0,
+     "required": true,
+     "description": "Text to narrate (max 10,000 characters). Use markers like <#0.5#> to insert pauses in seconds.",
+     "kind": "string",
+     "default": "",
+     "multiline": true
+    },
+    {
+     "name": "voice_id",
+     "order": 1,
+     "required": false,
+     "description": "Voice to synthesize. Pick any MiniMax system voice (e.g. English_Wiselady, English_Deep-VoicedGentleman) or a voice_id returned by https://replicate.com/minimax/voice-cloning. See the full list of voi",
+     "kind": "string",
+     "default": "English_Wiselady"
+    },
+    {
+     "name": "speed",
+     "order": 2,
+     "required": false,
+     "description": "Speech speed multiplier (0.5–2.0). Lower is slower, higher is faster.",
+     "kind": "float",
+     "default": 1,
+     "min": 0.5,
+     "max": 2
+    },
+    {
+     "name": "volume",
+     "order": 3,
+     "required": false,
+     "description": "Relative loudness. 1.0 is default MiniMax gain. Range 0–10.",
+     "kind": "float",
+     "default": 1,
+     "min": 0,
+     "max": 10
+    },
+    {
+     "name": "pitch",
+     "order": 4,
+     "required": false,
+     "description": "Semitone offset applied to the voice (−12 to +12).",
+     "kind": "int",
+     "default": 0,
+     "min": -12,
+     "max": 12
+    },
+    {
+     "name": "emotion",
+     "order": 5,
+     "required": false,
+     "description": "Desired delivery style. Use auto to let MiniMax choose, or pick a specific emotion.",
+     "kind": "combo",
+     "options": [
+      "auto",
+      "happy",
+      "sad",
+      "angry",
+      "fearful",
+      "disgusted",
+      "surprised",
+      "calm",
+      "fluent",
+      "neutral"
+     ],
+     "default": "auto"
+    },
+    {
+     "name": "english_normalization",
+     "order": 6,
+     "required": false,
+     "description": "Improve number/date reading for English text (adds a small amount of latency).",
+     "kind": "bool",
+     "default": false
+    },
+    {
+     "name": "sample_rate",
+     "order": 7,
+     "required": false,
+     "description": "Audio sample rate in Hz.",
+     "kind": "combo",
+     "options": [
+      8000,
+      16000,
+      22050,
+      24000,
+      32000,
+      44100
+     ],
+     "default": 32000
+    },
+    {
+     "name": "bitrate",
+     "order": 8,
+     "required": false,
+     "description": "MP3 bitrate in bits per second. Only used when audio_format is mp3.",
+     "kind": "combo",
+     "options": [
+      32000,
+      64000,
+      128000,
+      256000
+     ],
+     "default": 128000
+    },
+    {
+     "name": "audio_format",
+     "order": 9,
+     "required": false,
+     "description": "File format for the generated audio. Choose mp3 for general use, wav/flac for lossless, or pcm for raw bytes.",
+     "kind": "combo",
+     "options": [
+      "mp3",
+      "wav",
+      "flac",
+      "pcm"
+     ],
+     "default": "mp3"
+    },
+    {
+     "name": "channel",
+     "order": 10,
+     "required": false,
+     "description": "mono for 1 channel (default), stereo for 2 channels.",
+     "kind": "combo",
+     "options": [
+      "mono",
+      "stereo"
+     ],
+     "default": "mono"
+    },
+    {
+     "name": "subtitle_enable",
+     "order": 11,
+     "required": false,
+     "description": "Return MiniMax subtitle metadata with sentence timestamps (non-streaming only).",
+     "kind": "bool",
+     "default": false
+    },
+    {
+     "name": "language_boost",
+     "order": 12,
+     "required": false,
+     "description": "Optional language hint. Choose Automatic to let MiniMax detect the language, or pick a specific locale.",
+     "kind": "combo",
+     "options": [
+      "None",
+      "Automatic",
+      "Chinese",
+      "Chinese,Yue",
+      "Cantonese",
+      "English",
+      "Arabic",
+      "Russian",
+      "Spanish",
+      "French",
+      "Portuguese",
+      "German",
+      "Turkish",
+      "Dutch",
+      "Ukrainian",
+      "Vietnamese",
+      "Indonesian",
+      "Japanese",
+      "Italian",
+      "Korean",
+      "Thai",
+      "Polish",
+      "Romanian",
+      "Greek",
+      "Czech",
+      "Finnish",
+      "Hindi",
+      "Bulgarian",
+      "Danish",
+      "Hebrew",
+      "Malay",
+      "Persian",
+      "Slovak",
+      "Swedish",
+      "Croatian",
+      "Filipino",
+      "Hungarian",
+      "Norwegian",
+      "Slovenian",
+      "Catalan",
+      "Nynorsk",
+      "Tamil",
+      "Afrikaans"
+     ],
+     "default": "None"
+    }
+   ]
+  },
+  {
+   "id": "inworld/realtime-tts-2",
+   "name": "Inworld TTS 2",
+   "category": "tts",
+   "output": "audio",
+   "inputs": [
+    {
+     "name": "text",
+     "order": 0,
+     "required": true,
+     "description": "The text to convert to speech. Maximum 2,000 characters. Supports natural-language steering with bracketed instructions placed before the text they apply to (e.g. `[say excitedly]`, `[whisper in a hus",
+     "kind": "string",
+     "default": "",
+     "multiline": true
+    },
+    {
+     "name": "voice_id",
+     "order": 1,
+     "required": false,
+     "description": "The voice to use. Use a preset voice name (e.g. 'Ashley', 'Dennis', 'Alex', 'Darlene') or a custom cloned voice ID.",
+     "kind": "string",
+     "default": "Ashley"
+    },
+    {
+     "name": "language",
+     "order": 2,
+     "required": false,
+     "description": "Language of the input text. Use 'auto' to let the model detect the language. Supported production languages: English (en), Chinese (zh), Japanese (ja), Korean (ko), Russian (ru), Italian (it), Spanish",
+     "kind": "combo",
+     "options": [
+      "auto",
+      "en",
+      "zh",
+      "ja",
+      "ko",
+      "ru",
+      "it",
+      "es",
+      "pt",
+      "fr",
+      "de",
+      "pl",
+      "nl",
+      "hi",
+      "he",
+      "ar"
+     ],
+     "default": "auto"
+    },
+    {
+     "name": "temperature",
+     "order": 3,
+     "required": false,
+     "description": "Controls randomness when generating audio. Higher values produce more expressive results, lower values are more deterministic. Set to 0 to use the model default (1.1).",
+     "kind": "float",
+     "default": 0,
+     "min": 0,
+     "max": 2
+    },
+    {
+     "name": "audio_format",
+     "order": 4,
+     "required": false,
+     "description": "Output audio format.",
+     "kind": "combo",
+     "options": [
+      "mp3",
+      "wav",
+      "ogg_opus",
+      "flac"
+     ],
+     "default": "mp3"
+    },
+    {
+     "name": "sample_rate",
+     "order": 5,
+     "required": false,
+     "description": "Audio sample rate in Hz.",
+     "kind": "combo",
+     "options": [
+      8000,
+      16000,
+      22050,
+      24000,
+      32000,
+      44100,
+      48000
+     ],
+     "default": 48000
+    },
+    {
+     "name": "speaking_rate",
+     "order": 6,
+     "required": false,
+     "description": "Speaking speed multiplier. Set to 0 for normal speed (1.0).",
+     "kind": "float",
+     "default": 0,
+     "min": 0,
+     "max": 1.5
+    },
+    {
+     "name": "text_normalization",
+     "order": 7,
+     "required": false,
+     "description": "Controls whether numbers, dates, and abbreviations are expanded before synthesis. 'auto' lets the model decide, 'on' always normalizes, 'off' reads text as-is.",
+     "kind": "combo",
+     "options": [
+      "auto",
+      "on",
+      "off"
+     ],
+     "default": "auto"
+    }
+   ]
+  },
+  {
+   "id": "elevenlabs/v3",
+   "name": "ElevenLabs v3",
+   "category": "tts",
+   "output": "audio",
+   "inputs": [
+    {
+     "name": "prompt",
+     "order": 0,
+     "required": true,
+     "description": "The text to convert to speech",
+     "kind": "string",
+     "default": "",
+     "multiline": true
+    },
+    {
+     "name": "voice",
+     "order": 1,
+     "required": false,
+     "description": "Voice choice for speech generation",
+     "kind": "combo",
+     "options": [
+      "Rachel",
+      "Drew",
+      "Clyde",
+      "Paul",
+      "Aria",
+      "Domi",
+      "Dave",
+      "Roger",
+      "Fin",
+      "Sarah",
+      "James",
+      "Jane",
+      "Juniper",
+      "Arabella",
+      "Hope",
+      "Bradford",
+      "Reginald",
+      "Gaming",
+      "Austin",
+      "Kuon",
+      "Blondie",
+      "Priyanka",
+      "Alexandra",
+      "Monika",
+      "Mark",
+      "Grimblewood"
+     ],
+     "default": "Rachel"
+    },
+    {
+     "name": "stability",
+     "order": 2,
+     "required": false,
+     "description": "Stability setting for voice generation (0.0 to 1.0)",
+     "kind": "float",
+     "default": 0.5,
+     "min": 0,
+     "max": 1
+    },
+    {
+     "name": "similarity_boost",
+     "order": 3,
+     "required": false,
+     "description": "Similarity boost setting (0.0 to 1.0)",
+     "kind": "float",
+     "default": 0.75,
+     "min": 0,
+     "max": 1
+    },
+    {
+     "name": "style",
+     "order": 4,
+     "required": false,
+     "description": "Style exaggeration (0.0 to 1.0)",
+     "kind": "float",
+     "default": 0,
+     "min": 0,
+     "max": 1
+    },
+    {
+     "name": "speed",
+     "order": 5,
+     "required": false,
+     "description": "Speed of speech (0.25 to 4.0)",
+     "kind": "float",
+     "default": 1,
+     "min": 0.7,
+     "max": 1.2
+    },
+    {
+     "name": "previous_text",
+     "order": 6,
+     "required": false,
+     "description": "Previous text for context",
+     "kind": "string",
+     "default": "",
+     "multiline": true
+    },
+    {
+     "name": "next_text",
+     "order": 7,
+     "required": false,
+     "description": "Next text for context",
+     "kind": "string",
+     "default": "",
+     "multiline": true
+    },
+    {
+     "name": "language_code",
+     "order": 8,
+     "required": false,
+     "description": "Language code (e.g., 'en', 'es', 'fr')",
+     "kind": "string",
+     "default": "en"
+    }
+   ]
+  },
+  {
+   "id": "veed/fabric-1.0",
+   "name": "VEED Fabric 1.0",
+   "category": "lipsync",
+   "output": "video",
+   "inputs": [
+    {
+     "name": "image",
+     "order": 0,
+     "required": true,
+     "description": "Image to generate video from",
+     "kind": "image",
+     "is_list": false
+    },
+    {
+     "name": "audio",
+     "order": 1,
+     "required": true,
+     "description": "Audio to generate video from",
+     "kind": "audio"
+    },
+    {
+     "name": "resolution",
+     "order": 2,
+     "required": false,
+     "description": "Resolution of the generated video",
+     "kind": "combo",
+     "options": [
+      "480p",
+      "720p"
+     ],
+     "default": "720p"
+    }
+   ]
+  },
+  {
+   "id": "prunaai/p-video-avatar",
+   "name": "Pruna Video Avatar",
+   "category": "lipsync",
+   "output": "video",
+   "inputs": [
+    {
+     "name": "image",
+     "order": 0,
+     "required": true,
+     "description": "Input image (first frame). Supports jpg, jpeg, png, webp.",
+     "kind": "image",
+     "is_list": false
+    },
+    {
+     "name": "resolution",
+     "order": 1,
+     "required": false,
+     "description": "Resolution of the video.",
+     "kind": "combo",
+     "options": [
+      "720p",
+      "1080p"
+     ],
+     "default": "720p"
+    },
+    {
+     "name": "audio",
+     "order": 2,
+     "required": false,
+     "description": "Optional uploaded audio to drive avatar speech. If provided, this is used instead of voice_script and voice settings.",
+     "kind": "audio"
+    },
+    {
+     "name": "voice",
+     "order": 3,
+     "required": false,
+     "description": "Voice to use when generating speech from voice_script.",
+     "kind": "combo",
+     "options": [
+      "Zephyr (Female)",
+      "Puck (Male)",
+      "Charon (Male)",
+      "Kore (Female)",
+      "Fenrir (Male)",
+      "Leda (Female)",
+      "Orus (Male)",
+      "Aoede (Female)",
+      "Callirrhoe (Female)",
+      "Autonoe (Female)",
+      "Enceladus (Male)",
+      "Iapetus (Male)",
+      "Umbriel (Male)",
+      "Algenib (Male)",
+      "Despina (Female)",
+      "Erinome (Female)",
+      "Laomedeia (Female)",
+      "Achernar (Female)",
+      "Algieba (Male)",
+      "Schedar (Male)",
+      "Gacrux (Female)",
+      "Pulcherrima (Female)",
+      "Achird (Male)",
+      "Zubenelgenubi (Male)",
+      "Vindemiatrix (Female)",
+      "Sadachbia (Male)",
+      "Sadaltager (Male)",
+      "Sulafat (Female)",
+      "Alnilam (Male)",
+      "Rasalgethi (Male)"
+     ],
+     "default": "Zephyr (Female)"
+    },
+    {
+     "name": "voice_script",
+     "order": 4,
+     "required": false,
+     "description": "Exact words the avatar should say. Required when no audio file is uploaded.",
+     "kind": "string",
+     "default": "",
+     "multiline": true
+    },
+    {
+     "name": "voice_prompt",
+     "order": 5,
+     "required": false,
+     "description": "Optional style instructions for how to speak voice_script, such as tone, pacing, accent, or emotion. These instructions are not spoken.",
+     "kind": "string",
+     "default": "Say the following.",
+     "multiline": true
+    },
+    {
+     "name": "voice_language",
+     "order": 6,
+     "required": false,
+     "description": "Language/accent target for generated speech.",
+     "kind": "combo",
+     "options": [
+      "English (US)",
+      "English (UK)",
+      "Spanish",
+      "French",
+      "German",
+      "Italian",
+      "Portuguese (Brazil)",
+      "Japanese",
+      "Korean",
+      "Hindi"
+     ],
+     "default": "English (US)"
+    },
+    {
+     "name": "seed",
+     "order": 7,
+     "required": false,
+     "description": "Random seed. Set for reproducible generation.",
+     "kind": "int",
+     "default": 0,
+     "min": -2147483648,
+     "max": 2147483647
+    },
+    {
+     "name": "video_prompt",
+     "order": 8,
+     "required": false,
+     "description": "Optional visual prompt describing how the person should appear or behave while speaking.",
+     "kind": "string",
+     "default": "The person is talking.",
+     "multiline": true
+    },
+    {
+     "name": "negative_prompt",
+     "order": 9,
+     "required": false,
+     "description": "Disabled if empty.Mention what you do NOT want in the video, e.g. \"subtitles, text, blurry, low quality, frames, watermark, titles, scene change\". We recommend using multiple keywords at once.",
+     "kind": "string",
+     "default": "",
+     "multiline": true
+    },
+    {
+     "name": "strength_negative_prompt",
+     "order": 10,
+     "required": false,
+     "description": "Strength of the Negative Prompt. Optimal value can differ for different video lengths (Experimental Feature)",
+     "kind": "float",
+     "default": 0.5,
+     "min": 0,
+     "max": 4
+    },
+    {
+     "name": "disable_safety_filter",
+     "order": 11,
+     "required": false,
+     "description": "Disable safety filter for prompts and input image. When disabled, prompts are not checked for unsafe content before generation.",
+     "kind": "bool",
+     "default": true
+    },
+    {
+     "name": "disable_prompt_upsampling",
+     "order": 12,
+     "required": false,
+     "description": "When true, skip automatic enhancement of the visual video prompt and use video_prompt directly.",
+     "kind": "bool",
+     "default": false
+    },
+    {
+     "name": "no_op",
+     "order": 13,
+     "required": false,
+     "description": "Health check mode - returns status without inference.",
+     "kind": "bool",
+     "default": false
+    }
+   ]
+  },
+  {
+   "id": "sync/lipsync-2-pro",
+   "name": "Sync Lipsync 2 Pro",
+   "category": "lipsync",
+   "output": "video",
+   "inputs": [
+    {
+     "name": "video",
+     "order": 0,
+     "required": true,
+     "description": "Input video file (.mp4)",
+     "kind": "uri_string"
+    },
+    {
+     "name": "audio",
+     "order": 1,
+     "required": true,
+     "description": "Input audio file (.wav)",
+     "kind": "audio"
+    },
+    {
+     "name": "sync_mode",
+     "order": 2,
+     "required": false,
+     "description": "Lipsync mode when audio and video durations are out of sync",
+     "kind": "combo",
+     "options": [
+      "loop",
+      "bounce",
+      "cut_off",
+      "silence",
+      "remap"
+     ],
+     "default": "loop"
+    },
+    {
+     "name": "temperature",
+     "order": 3,
+     "required": false,
+     "description": "How expressive lipsync can be (0-1)",
+     "kind": "float",
+     "default": 0.5,
+     "min": 0,
+     "max": 1
+    },
+    {
+     "name": "active_speaker",
+     "order": 4,
+     "required": false,
+     "description": "Whether to detect active speaker (i.e. whoever is speaking in the clip will be used for lipsync)",
+     "kind": "bool",
+     "default": false
+    }
+   ]
+  },
+  {
+   "id": "heygen/lipsync-precision",
+   "name": "HeyGen Lipsync Precision",
+   "category": "lipsync",
+   "output": "video",
+   "inputs": [
+    {
+     "name": "video",
+     "order": 0,
+     "required": true,
+     "description": "Source video file to lip-sync.",
+     "kind": "uri_string"
+    },
+    {
+     "name": "audio",
+     "order": 1,
+     "required": true,
+     "description": "Replacement audio file. The video's lip movements will be re-animated to match this audio using high-accuracy avatar inference.",
+     "kind": "audio"
+    },
+    {
+     "name": "enable_dynamic_duration",
+     "order": 2,
+     "required": false,
+     "description": "Allow the output duration to adjust to match the new audio length.",
+     "kind": "bool",
+     "default": true
+    },
+    {
+     "name": "disable_music_track",
+     "order": 3,
+     "required": false,
+     "description": "Strip background music from the source video.",
+     "kind": "bool",
+     "default": false
+    },
+    {
+     "name": "enable_speech_enhancement",
+     "order": 4,
+     "required": false,
+     "description": "Enhance speech quality in the output.",
+     "kind": "bool",
+     "default": false
+    }
+   ]
+  }
+ ]
+}
\ No newline at end of file
diff --git a/nodes.py b/nodes.py
new file mode 100644
index 0000000..0852c3a
--- /dev/null
+++ b/nodes.py
@@ -0,0 +1,262 @@
+"""ComfyUI-Replicate-Select — kuratierte Replicate-Modelle als ComfyUI-Nodes.
+
+Architektur nach Runware-Vorbild: nodes.json (Katalog) + Generator + Custom-Node.
+Auth: REPLICATE_API_TOKEN aus der Umgebung (Modal-Secret mcp-keys).
+Katalog aktualisieren: python scripts/build_catalog.py
+"""
+import base64
+import io
+import json
+import os
+import time
+from pathlib import Path
+
+import numpy as np
+import torch
+
+CATALOG_PATH = Path(__file__).parent / "nodes.json"
+CATEGORY_ROOT = "Replicate Select"
+DATA_URI_LIMIT = 200 * 1024  # größere Dateien gehen über die Files-API
+
+
+# ---------- Replicate-Client ----------
+
+def _client():
+    import replicate
+    if not os.environ.get("REPLICATE_API_TOKEN"):
+        raise RuntimeError(
+            "REPLICATE_API_TOKEN fehlt. In Modal: Secret 'mcp-keys' prüfen; "
+            "lokal: export REPLICATE_API_TOKEN=r8_..."
+        )
+    return replicate
+
+
+def _upload_bytes(data: bytes, filename: str, content_type: str) -> str:
+    """Datei zu Replicate hochladen (Files-API) oder als data-URI inline."""
+    if len(data) <= DATA_URI_LIMIT:
+        b64 = base64.b64encode(data).decode()
+        return f"data:{content_type};base64,{b64}"
+    import replicate
+    f = replicate.files.create(io.BytesIO(data), {"filename": filename, "content_type": content_type})
+    return f.urls["get"]
+
+
+# ---------- Medien-Konvertierung ----------
+
+def _image_to_uri(image_tensor) -> str:
+    """ComfyUI IMAGE (B,H,W,C float 0-1) → PNG-URI (erstes Bild im Batch)."""
+    from PIL import Image
+    arr = (image_tensor[0].cpu().numpy() * 255).clip(0, 255).astype(np.uint8)
+    img = Image.fromarray(arr)
+    buf = io.BytesIO()
+    img.save(buf, format="PNG")
+    return _upload_bytes(buf.getvalue(), "input.png", "image/png")
+
+
+def _audio_to_uri(audio: dict) -> str:
+    """ComfyUI AUDIO ({'waveform': (B,C,T), 'sample_rate': int}) → WAV-URI."""
+    import torchaudio
+    buf = io.BytesIO()
+    waveform = audio["waveform"][0]  # (C, T)
+    torchaudio.save(buf, waveform.cpu(), audio["sample_rate"], format="wav")
+    return _upload_bytes(buf.getvalue(), "input.wav", "audio/wav")
+
+
+def _download(url_or_file) -> bytes:
+    """Replicate-Output (FileOutput oder URL-String) → Bytes."""
+    if hasattr(url_or_file, "read"):
+        return url_or_file.read()
+    import urllib.request
+    with urllib.request.urlopen(str(url_or_file), timeout=600) as resp:
+        return resp.read()
+
+
+def _to_image_tensor(data: bytes):
+    from PIL import Image
+    img = Image.open(io.BytesIO(data)).convert("RGB")
+    arr = np.asarray(img).astype(np.float32) / 255.0
+    return torch.from_numpy(arr)[None,]
+
+
+def _to_audio(data: bytes) -> dict:
+    import torchaudio
+    waveform, sr = torchaudio.load(io.BytesIO(data))
+    return {"waveform": waveform[None,], "sample_rate": sr}
+
+
+def _save_video(data: bytes, model_id: str) -> str:
+    """Video in den ComfyUI-Output-Ordner schreiben, Pfad zurückgeben."""
+    try:
+        import folder_paths
+        out_dir = Path(folder_paths.get_output_directory())
+    except Exception:
+        out_dir = Path("output")
+    out_dir.mkdir(parents=True, exist_ok=True)
+    slug = model_id.replace("/", "_")
+    path = out_dir / f"replicate_{slug}_{int(time.time())}.mp4"
+    path.write_bytes(data)
+    return str(path)
+
+
+def _first_output(result):
+    """Replicate gibt je nach Modell FileOutput, Liste oder dict zurück."""
+    if isinstance(result, (list, tuple)):
+        return result[0] if result else None
+    if isinstance(result, dict):
+        for key in ("video", "audio", "image", "output", "url"):
+            if key in result:
+                return result[key]
+        return next(iter(result.values()), None)
+    return result
+
+
+# ---------- Node-Generierung ----------
+
+def _comfy_input(entry: dict):
+    kind = entry["kind"]
+    tooltip = {"tooltip": entry["description"]} if entry.get("description") else {}
+    if kind == "combo":
+        return ([str(o) for o in entry["options"]], {"default": str(entry.get("default", "")), **tooltip})
+    if kind == "bool":
+        return ("BOOLEAN", {"default": entry.get("default", False), **tooltip})
+    if kind == "int":
+        return ("INT", {"default": int(entry.get("default") or 0),
+                        "min": int(max(entry.get("min", -2**31), -2**31)),
+                        "max": int(min(entry.get("max", 2**31 - 1), 2**31 - 1)), **tooltip})
+    if kind == "float":
+        return ("FLOAT", {"default": float(entry.get("default") or 0.0),
+                          "min": float(entry.get("min", -1e9)), "max": float(entry.get("max", 1e9)),
+                          "step": 0.01, **tooltip})
+    if kind == "image":
+        return ("IMAGE", tooltip)
+    if kind == "audio":
+        return ("AUDIO", tooltip)
+    # string / uri_string
+    return ("STRING", {"default": str(entry.get("default") or ""),
+                       "multiline": entry.get("multiline", False), **tooltip})
+
+
+def _build_arguments(model_def: dict, kwargs: dict) -> dict:
+    args = {}
+    for entry in model_def["inputs"]:
+        name = entry["name"]
+        value = kwargs.get(name)
+        if value is None:
+            continue
+        kind = entry["kind"]
+        if kind == "image":
+            uri = _image_to_uri(value)
+            args[name] = [uri] if entry.get("is_list") else uri
+        elif kind == "audio":
+            args[name] = _audio_to_uri(value)
+        elif kind == "combo":
+            args[name] = value
+        elif kind in ("string", "uri_string"):
+            if str(value).strip():
+                args[name] = value
+        else:
+            args[name] = value
+    return args
+
+
+_RETURN_BY_OUTPUT = {
+    "image": (("IMAGE",), ("image",)),
+    "audio": (("AUDIO",), ("audio",)),
+    "video": (("STRING",), ("video_path",)),
+}
+
+
+def _make_node(model_def: dict):
+    output_kind = model_def["output"]
+    return_types, return_names = _RETURN_BY_OUTPUT[output_kind]
+
+    required, optional = {}, {}
+    for entry in model_def["inputs"]:
+        target = required if entry["required"] else optional
+        target[entry["name"]] = _comfy_input(entry)
+
+    class ReplicateModelNode:
+        CATEGORY = f"{CATEGORY_ROOT}/{model_def['category']}"
+        FUNCTION = "run"
+        RETURN_TYPES = return_types
+        RETURN_NAMES = return_names
+
+        @classmethod
+        def INPUT_TYPES(cls):
+            spec = {"required": dict(required)}
+            if optional:
+                spec["optional"] = dict(optional)
+            return spec
+
+        def run(self, **kwargs):
+            replicate = _client()
+            args = _build_arguments(model_def, kwargs)
+            print(f"[Replicate Select] {model_def['id']} ← {list(args.keys())}")
+            result = replicate.run(model_def["id"], input=args)
+            output = _first_output(result)
+            if output is None:
+                raise RuntimeError(f"{model_def['id']}: leere Antwort")
+            data = _download(output)
+            if output_kind == "image":
+                return (_to_image_tensor(data),)
+            if output_kind == "audio":
+                return (_to_audio(data),)
+            return (_save_video(data, model_def["id"]),)
+
+    ReplicateModelNode.__name__ = "Replicate_" + model_def["id"].replace("/", "_").replace("-", "_").replace(".", "_")
+    return ReplicateModelNode
+
+
+class ReplicateCustom:
+    """Beliebiges Replicate-Modell per ID aufrufen — der Escape-Hatch."""
+
+    CATEGORY = CATEGORY_ROOT
+    FUNCTION = "run"
+    RETURN_TYPES = ("STRING", "IMAGE")
+    RETURN_NAMES = ("raw_output", "image")
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "model_id": ("STRING", {"default": "owner/model", "tooltip": "z.B. bytedance/seedream-4.5"}),
+                "input_json": ("STRING", {"multiline": True, "default": '{\n  "prompt": ""\n}',
+                                          "tooltip": "Input-Parameter als JSON"}),
+            },
+            "optional": {
+                "image": ("IMAGE", {"tooltip": "Wird als 'image'-Parameter angehängt, falls gesetzt"}),
+            },
+        }
+
+    def run(self, model_id, input_json, image=None):
+        replicate = _client()
+        args = json.loads(input_json)
+        if image is not None:
+            args["image"] = _image_to_uri(image)
+        result = replicate.run(model_id.strip(), input=args)
+        output = _first_output(result)
+        raw = str(output)
+        image_out = torch.zeros(1, 1, 1, 3)
+        try:
+            data = _download(output)
+            if data[:8].startswith((b"\x89PNG", b"\xff\xd8")):
+                image_out = _to_image_tensor(data)
+            else:
+                raw = f"{raw}\n(bytes: {len(data)})"
+        except Exception:
+            pass
+        return (raw, image_out)
+
+
+def _load_catalog() -> dict:
+    with open(CATALOG_PATH, encoding="utf-8") as fh:
+        return json.load(fh)
+
+
+NODE_CLASS_MAPPINGS = {"ReplicateCustom": ReplicateCustom}
+NODE_DISPLAY_NAME_MAPPINGS = {"ReplicateCustom": "Replicate Custom (beliebige Modell-ID)"}
+
+for _model in _load_catalog()["models"]:
+    _cls = _make_node(_model)
+    NODE_CLASS_MAPPINGS[_cls.__name__] = _cls
+    NODE_DISPLAY_NAME_MAPPINGS[_cls.__name__] = f"{_model['name']} (Replicate)"
diff --git a/pyproject.toml b/pyproject.toml
new file mode 100644
index 0000000..8e7a262
--- /dev/null
+++ b/pyproject.toml
@@ -0,0 +1,9 @@
+[project]
+name = "comfyui-replicate-select"
+description = "Kuratierte Replicate-Modelle als ComfyUI-Nodes (Bild, Video, Upscale, TTS, Lipsync) + Custom-Node für beliebige Modell-IDs."
+version = "0.1.0"
+dependencies = ["replicate>=1.0.7"]
+
+[tool.comfy]
+PublisherId = "pharma-lobby"
+DisplayName = "Replicate Select"
diff --git a/requirements.txt b/requirements.txt
new file mode 100644
index 0000000..849fe83
--- /dev/null
+++ b/requirements.txt
@@ -0,0 +1 @@
+replicate>=1.0.7
diff --git a/scripts/build_catalog.py b/scripts/build_catalog.py
new file mode 100644
index 0000000..0268647
--- /dev/null
+++ b/scripts/build_catalog.py
@@ -0,0 +1,140 @@
+#!/usr/bin/env python3
+"""Baut nodes.json aus den öffentlichen Replicate-Schema-Seiten (kein API-Token nötig).
+
+Aufruf:  python scripts/build_catalog.py
+Schreibt: nodes.json im Repo-Root.
+
+Neue Modelle: unten in META eintragen, Script laufen lassen, fertig.
+"""
+import json
+import re
+import time
+import urllib.request
+from pathlib import Path
+
+# Modell-ID → (Anzeigename, Kategorie, Output-Art: image|video|audio)
+META = {
+    "bytedance/seedream-4.5":      ("Seedream 4.5",         "image",   "image"),
+    "google/nano-banana-pro":      ("Nano Banana Pro",      "image",   "image"),
+    "kwaivgi/kling-v3-video":      ("Kling v3 Video",       "video",   "video"),
+    "google/veo-3.1":              ("Veo 3.1",              "video",   "video"),
+    "bytedance/seedance-2.0-fast": ("Seedance 2.0 Fast",    "video",   "video"),
+    "cjwbw/supir":                 ("SUPIR Upscale",        "upscale", "image"),
+    "cjwbw/real-esrgan":           ("Real-ESRGAN Upscale",  "upscale", "image"),
+    "minimax/speech-2.8-hd":       ("MiniMax Speech 2.8 HD","tts",     "audio"),
+    "inworld/realtime-tts-2":      ("Inworld TTS 2",        "tts",     "audio"),
+    "elevenlabs/v3":               ("ElevenLabs v3",        "tts",     "audio"),
+    "veed/fabric-1.0":             ("VEED Fabric 1.0",      "lipsync", "video"),
+    "prunaai/p-video-avatar":      ("Pruna Video Avatar",   "lipsync", "video"),
+    "sync/lipsync-2-pro":          ("Sync Lipsync 2 Pro",   "lipsync", "video"),
+    "heygen/lipsync-precision":    ("HeyGen Lipsync Precision", "lipsync", "video"),
+}
+
+
+def fetch_schema(model_id: str) -> dict | None:
+    """Schema aus dem eingebetteten JSON der öffentlichen Schema-Seite ziehen."""
+    req = urllib.request.Request(
+        f"https://replicate.com/{model_id}/api/schema",
+        headers={"User-Agent": "Mozilla/5.0"},
+    )
+    html = urllib.request.urlopen(req, timeout=30).read().decode()
+    for m in re.finditer(r'<script[^>]*type="application/json"[^>]*>(.*?)</script>', html, re.DOTALL):
+        try:
+            data = json.loads(m.group(1))
+        except Exception:
+            continue
+        version = data.get("version") or data
+        schema = version.get("_extras", {}).get("dereferenced_openapi_schema") or {}
+        schemas = schema.get("components", {}).get("schemas", {})
+        if schemas.get("Input"):
+            return {"input": schemas["Input"], "version_id": version.get("id")}
+    return None
+
+
+def classify_uri(name: str, is_list: bool) -> dict:
+    """URI-Inputs nach Medientyp einsortieren."""
+    lname = name.lower()
+    if any(k in lname for k in ("audio", "voice_sample", "speech")):
+        return {"kind": "audio"}
+    if any(k in lname for k in ("video", "motion")):
+        return {"kind": "uri_string"}  # Video-Inputs: URL als String (zu groß für Upload-Tricks)
+    if any(k in lname for k in ("image", "img", "frame", "mask", "reference")):
+        return {"kind": "image", "is_list": is_list}
+    return {"kind": "uri_string"}
+
+
+def distill(model_id: str, input_schema: dict) -> dict:
+    props = input_schema.get("properties", {})
+    required = set(input_schema.get("required", []))
+    inputs = []
+    for name, p in props.items():
+        if "allOf" in p and p["allOf"]:
+            merged = dict(p["allOf"][0])
+            merged.update({k: v for k, v in p.items() if k != "allOf"})
+            p = merged
+        entry = {
+            "name": name,
+            "order": p.get("x-order", 999),
+            "required": name in required,
+            "description": (p.get("description") or "")[:200],
+        }
+        t = p.get("type")
+        fmt = p.get("format", "")
+        items = p.get("items", {}) if isinstance(p.get("items"), dict) else {}
+        is_uri = fmt == "uri" or items.get("format") == "uri"
+        is_list = t == "array"
+        lname = name.lower()
+
+        if p.get("enum"):
+            entry["kind"] = "combo"
+            entry["options"] = p["enum"]
+            entry["default"] = p.get("default", p["enum"][0])
+        elif t == "boolean":
+            entry["kind"] = "bool"
+            entry["default"] = bool(p.get("default", False))
+        elif t == "integer":
+            entry["kind"] = "int"
+            entry["default"] = p.get("default", 0)
+            entry["min"] = p.get("minimum", -2**31)
+            entry["max"] = p.get("maximum", 2**31 - 1)
+        elif t == "number":
+            entry["kind"] = "float"
+            entry["default"] = p.get("default", 0.0)
+            entry["min"] = p.get("minimum", -1e9)
+            entry["max"] = p.get("maximum", 1e9)
+        elif is_uri:
+            entry.update(classify_uri(name, is_list))
+        else:
+            entry["kind"] = "string"
+            entry["default"] = p.get("default") or ""
+            if any(k in lname for k in ("prompt", "text", "instruction", "script")):
+                entry["multiline"] = True
+        inputs.append(entry)
+    inputs.sort(key=lambda e: (not e["required"], e["order"]))
+    name, cat, out = META[model_id]
+    return {"id": model_id, "name": name, "category": cat, "output": out, "inputs": inputs}
+
+
+def main():
+    models = []
+    for model_id in META:
+        try:
+            fetched = fetch_schema(model_id)
+        except Exception as exc:
+            print(f"✗ {model_id}: {exc}")
+            continue
+        if not fetched:
+            print(f"✗ {model_id}: kein Schema gefunden")
+            continue
+        entry = distill(model_id, fetched["input"])
+        models.append(entry)
+        print(f"✓ {model_id}: {len(entry['inputs'])} inputs")
+        time.sleep(1)
+
+    out_path = Path(__file__).resolve().parent.parent / "nodes.json"
+    out_path.write_text(json.dumps({"models": models}, indent=1, ensure_ascii=False))
+    print(f"→ {out_path} ({len(models)}/{len(META)} Modelle)")
+
+
+if __name__ == "__main__":
+    main()