Multimodal Genai Standards

Generative and understanding systems over non-text modalities — image, audio, video and document — as an engineering problem, provider-agnostic. Use when sending an image, a PDF page, a screenshot, a chart, an audio file or a video frame into a vision-language model, choosing between a classical OCR pipeline (Tesseract, PaddleOCR, docTR, Surya/Marker) and a VLM-based document reader (dots.ocr, olmOCR, DeepSeek-OCR, PaddleOCR-VL, Docling) and benchmarking with OmniDocBench or olmOCR-Bench, generating images with diffusion models (Stable Diffusion, FLUX.1 dev/schnell/pro, ComfyUI, ControlNet, LoRA adapters, seed/CFG/steps, negative prompts) and budgeting cost per image, running speech-to-text (Whisper large-v3 and turbo, faster-whisper, whisper.cpp, Parakeet TDT, Canary-Qwen, Voxtral, WER measurement, diarization, timestamps) or text-to-speech and voice cloning consent, processing video (frame sampling rate, keyframes, temporal cost), estimating latency and cost per modality and per token of image or audio, eva

serialexperimentslainnnn Updated 1 repo stars

File contents

serialexperimentslainnnn/claude-code-native-skills-and-settings commit a8686508c7

Frequently asked questions

npx get-skill add serialexperimentslainnnn/multimodal-genai-standards