Ingredient & Recipe Catalog

(Part 1 of 4 — same chapter in the PDF; split for the web site.)

Appendix B Ingredient & Recipe Catalog This appendix is generated from the same source as planning/INGREDIENT_RECIPE_CATALOG.md: PostgreSQL ingredient registry (when available), optional ingredients//avb.json (kind: node) and recipes/.avb.json (kind: graph) overlays (docs/AVB_JSON.md), plus recipes/.json. Regenerate before building the PDF: python scripts/generate_ingredient_recipe_catalog.py —format latex Or use make -C docs/whitepaper/latex pdf (runs the generator, then latexmk). CI without a database may use –recipes-only (ingredient Agent nodes section stubbed; Agent graphs still listed). The generated body uses one \subsection* per entry, with fields in a description list (enumitem, style=nextline) so labels sit above values; slugs and paths use breakable \path (xurl). Recipe-type rows in Agent nodes reuse recipes/*.avb.json and fall back to metadata.description from the recipe JSON when no summary is in AVB. Agent nodes (ingredients) Each block lists slug, usageClass, and summary (from ingredients//avb.json when present; otherwise name/class fall back to the registry and heuristics). Optional httpPath / publicBaseUrl appear when set. See docs/AVB_JSON.md. AWS Bedrock LLM Provider Slug aws_bedrock_llm Usage class AWS / Bedrock Summary Generic AWS Bedrock text completion with support for Claude, Titan, and Llama model families, flexible authentication, and standardized cost reporting Code path (guess) ingredients/aws_bedrock_llm/ FAL.ai Authentication Provider Slug fal_ai Usage class other / core 91

Summary The FAL.ai ingredient serves as the central point for all FAL.ai external API interactions in the ScryptedAI platform. It handles authentication, job submission, status polling, and result retrieval for various FAL.ai services including image generation, video generation, and audio generation. Code path (guess) ingredients/fal_ai/ ElevenLabs Dialogue v3 Text-to-Dialogue Slug fal:elevenlabs-dialogue-v3:text-to-dialogue Usage class provider (FAL) Summary ElevenLabs Dialogue v3 ingredient handler. This ingredient provides a provider-agnostic interface for ElevenLabs Dialogue v3 text-to-dialogue generation. Currently uses FAL.ai as the default provider. Hailuo 2.3 Image-to-Video Slug fal:hailuo-2-3-i2v:image-to-video Usage class provider (FAL) Summary Hailuo 2.3 Image-to-Video ingredient handler. This ingredient provides a provider-agnostic interface for generating videos from images using Hailuo 2.3 model. Currently uses FAL.ai as the default provider. Kling Video 2.5 Turbo Pro I2V Slug fal:kling-video:i2v Usage class provider (FAL) Summary Generates video from a static image using Kling Video 2.5 Turbo Pro model via FAL.ai. Takes a source image URL and a motion prompt to create animated video content. MMAudio v2 - Video Soundtrack Generation Slug fal:mmaudio-v2 Usage class provider (FAL) Summary Add AI-generated soundtracks to videos by analyzing visual content and responding to text prompts. Perfect for adding background music, sound effects, or contextual audio to silent videos up to 29.5 seconds long. 92

Code path (guess) ingredients/mmaudio_v2/ Nano-Banana Edit Image Slug fal:nano-banana-edit:text-to-image Usage class provider (FAL) Summary Nano-Banana Edit Image ingredient handler. This ingredient provides a provider-agnostic interface for nano-banana image editing. Takes multiple images and a prompt to create a new edited image. Currently uses FAL.ai as the default provider. Nano-Banana Pro Edit Image Slug fal:nano-banana-pro-edit:text-to-image Usage class provider (FAL) Summary Nano-Banana Pro Edit Image ingredient handler. This ingredient provides a provider-agnostic interface for nano-banana-pro image editing. Takes multiple images and a prompt to create a new edited image with resolution support. Currently uses FAL.ai as the default provider. Nano-Banana Pro Text-to-Image Slug fal:nano-banana-pro:text-to-image Usage class provider (FAL) Summary Nano-Banana Pro Text-to-Image ingredient handler. This ingredient provides a provider-agnostic interface for nano-banana-pro text-to-image generation. Currently uses FAL.ai as the default provider, but can be extended to support other providers. Nano-Banana Text-to-Image Slug fal:nano-banana:text-to-image Usage class provider (FAL) Summary Nano-Banana Text-to-Image ingredient handler. This ingredient provides a provider-agnostic interface for nano-banana text-to-image generation. Currently uses FAL.ai as the default provider, but can be extended to support other providers. 93

OVI Image-to-Video Slug fal:ovi-image-to-video Usage class provider (FAL) Summary This ingredient generates videos from images using the OVI model via FAL.ai. Code path (guess) ingredients/ovi_image_to_video/ Seedream 4.0 Image Edit Slug fal:seedream-4:edit Usage class provider (FAL) Summary The seedream-edit ingredient enables editing of existing images based on text prompts. This is a key component for the character/scene registry toolchain, allowing extraction and modification of characters or scenes from existing images. Code path (guess) ingredients/seedream/ Seedream 4.0 Image Edit Slug fal:seedream-4:text-to-image Usage class provider (FAL) Summary Seedream 4.0 Text-to-Image ingredient handler. This ingredient provides a provider-agnostic interface for Seedream 4.0 text-to-image generation. Currently uses FAL.ai as the default provider, but can be extended to support other providers. Code path (guess) ingredients/seedream/ Text-to-Image with Storage Slug fal:text-to-image-with-storage Usage class other / core Summary Text-to-Image with Storage recipe handler. This ingredient provides a provider-agnostic interface for text-to-image generation with storage. Currently uses FAL.ai as the default provider, but can be extended to support other providers. 94

Code path (guess) ingredients/text_to_image_with_storage/ Vision Analysis Structured Slug scrypted:analysis:vision-structured Usage class analysis / vision Summary The Vision Analysis Structured ingredient provides comprehensive vision analysis for both images and videos, returning structured data with official style and motion type classifications. It leverages AWS Nova Vision models with sophisticated prompting to ensure consistent, structured output using only predefined classification tags. 1 Slug scrypted:analyzer:delula-prompt Usage class other / core Summary Delula Prompt Analyzer Ingredient Handler. Analyzes plain intent prompts and extracts structured information for recipe configuration, including audio requirements, optimized prompts, and speech data using AWS Nova Premier. Video Motion Estimator Slug scrypted:analyzer:video-motion Usage class other / core Summary The Video Motion Estimator analyzes videos to determine the motion rate of objects within the scene, normalized as “frame diagonal rate per second”. This powerful primitive uses sophisticated computer vision techniques to distinguish between camera motion (pan, rotation, zoom) and actual object motion, providing a clean metric for video activity analysis. Morpheus API Gateway Slug scrypted:api:morpheus Usage class external API bridge Summary The morpheus_api ingredient provides access to the Morpheus API Gateway (https://api.mor.org/api/v1 ) for chat completions, embeddings, models, and other AI marketplace services. It uses the local_credentials ingredient to securely fetch the API key from the encrypted keyring. 95

AWS Bedrock Guardrails Slug scrypted:aws:bedrock-guardrails Usage class moderation / safety Summary Content moderation and safety filtering using Amazon Bedrock Guardrails for text and image analysis with configurable safety policies AWS Credentials Provider Slug scrypted:aws:credentials Usage class AWS / Bedrock Summary AWS credentials ingredient handler - provides authenticated AWS session for other ingredients to use. Code path (guess) ingredients/aws_credentials/ AspectRatio Converter Slug scrypted:converter:aspect-ratio Usage class converter / util Summary Converts aspect ratio names (portrait, landscape, square) to pixel dimensions optimized for Seedream image generation. Audio Prompt to Flag Slug scrypted:converter:audio-prompt-to-flag Usage class converter / util Summary Audio Prompt to Flag Converter. Determines if audio should be generated and combines video and audio prompts. If audio_prompt is empty, sets generate_audio=False and uses video_prompt as-is. If audio_prompt has content, sets generate_audio=True and combines prompts. Code path (guess) ingredients/audio_prompt_to_flag/ 96

ELI5 Title Cleaner Slug scrypted:converter:eli5-title-cleaner Usage class converter / util Summary Cleans and repairs ELI5 (Explain Like I’m 5) article titles by fixing spelling mistakes, grammar, punctuation, removing unnecessary prefixes, and making titles unique. Also provides confidence scores indicating whether titles are valid ELI5 questions. Code path (guess) ingredients/eli5_title_cleaner/ Orientation to Canvas Dimensions Converter Slug scrypted:converter:orientation-to-canvas-dimensions Usage class converter / util Summary Orientation to Canvas Dimensions Converter. Converts orientation names (portrait, landscape, square) to pixel dimensions optimized for AWS Nova Canvas (1024-2048, multiples of 128). Code path (guess) ingredients/orientation_to_canvas_dimensions/ Orientation to Crop Aspect Ratio Converter Slug scrypted:converter:orientation-to-crop-aspect Usage class converter / util Summary Orientation to Crop Aspect Ratio Converter. Converts orientation intent to aspect ratio mapping (16:9, 1:1, 9:16). Returns null for landscape since NovaReel already outputs 16:9 and doesn’t need cropping. Code path (guess) ingredients/orientation_to_crop_aspect/ Orientation to Veo Aspect Ratio Slug scrypted:converter:orientation-to-veo-aspect-ratio Usage class converter / util Summary Orientation to Veo Aspect Ratio Converter. Converts orientation names (portrait, square, landscape) to Google Veo 3.1 Fast aspect ratios (9:16, 1:1, 16:9). 97

Code path (guess) ingredients/orientation_to_veo_aspect_ratio/ Delula Prompt Analyzer Slug scrypted:delula:prompt-analyzer Usage class other / core Summary Analyze plain intent prompts and extract structured information for recipe configuration, including audio requirements, optimized prompts, and speech data HTTP path POST /api/admin/recipe-creator/analyze BGE-M3 Embeddings Slug scrypted:embedding:bge-m3 Usage class embeddings Summary BGE-M3 Embeddings ingredient handler. This ingredient provides access to the BGE-M3 (BAAI General Embedding) model via the Morpheus API Gateway. BGE-M3 is a multilingual embedding model that generates 1024-dimensional dense vectors for semantic search, retrieval-augmented generation (RAG), and vector database applications. First URL Extractor Slug scrypted:extractor:first-url Usage class other / core Summary First URL Extractor ingredient handler - extracts first URL from array of image/video objects. Video Frames Extractor Slug scrypted:extractor:video-frames Usage class other / core Summary Extracts one or multiple frames from video files at specified frame indices using ffmpeg. Supports efficient batch extraction to minimize video loading overhead and stores extracted frames in the CDN with proper asset retention management. 98

Forecast Sundial Slug scrypted:forecast:sundial Usage class other / core Summary Sundial Forecasting ingredient handler. This ingredient provides time-series forecasting using the Sundial foundation model. It handles normalization of input data (100-100k points) to exactly 2880 points, calls the external Sundial API, and expands results based on the normalization factor. Code path (guess) ingredients/forecast_sundial/ AWS Nova Canvas (Bedrock) Slug scrypted:image:aws-nova-canvas Usage class generation (image) Summary Text-to-image via Amazon Nova Canvas on AWS Bedrock (InvokeModel). Supports 1024–2048 px (steps of 128), up to 1000-character prompts, 1–4 images per call, optional seed; returns base64 image data. Synchronous provider path; orchestration wraps it like other Bedrock ingredients. Backfill as Square Slug scrypted:image:backfill-as-square Usage class other / core Summary Backfill-as-Square Ingredient - Extends a non-square image to square by padding with average background color. Code path (guess) ingredients/backfill_as_square/ FLUX 2 Pro Text-to-Image Slug scrypted:image:flux-2-pro Usage class other / core Summary FLUX 2 Pro Text-to-Image ingredient handler. Uses FAL.ai fal-ai/flux-2-pro for high-quality text-to-image generation. Accepts the same orientation vocabulary as the rest of the API (portrait / landscape / square) and maps to FAL image_size; 16:9 and 9:16 use 1920x1080 and 1080x1920 per product requirement. 99

Grok Imagine Image Text-to-Image Slug scrypted:image:grok-imagine-image Usage class other / core Summary Grok Imagine Image text-to-image ingredient handler. Uses xAI’s Grok Imagine Image model via FAL.ai (xai/grok-imagine-image). Supports aspect_ratio, num_images, output_format (jpeg/png/webp), sync_mode. Code path (guess) ingredients/grok_imagine_image/ Image Resize Slug scrypted:image:resize Usage class other / core Summary Image Resize Ingredient - Resizes images to specific dimensions. Code path (guess) ingredients/image_resize/ Image Inspector Slug scrypted:inspector:image Usage class analysis / vision Summary Ingredient ID: scrypted:inspector:image Video Inspector Slug scrypted:inspector:video Usage class analysis / vision Summary The Video Inspector ingredient (scrypted:inspector:video) analyzes video files to extract comprehensive metadata including dimensions, duration, framerate, codec information, and color properties. Intent Interpreter Slug scrypted:interpreter:intent 100

Source: transcribed from the compiled Scrypted Network Design whitepaper PDF for web reading. Layout, figures, and pagination may differ from the PDF.