skills.rest
NVIDIA
NVIDIAOFFICIAL
Explore all skills by this author →
WebsiteGitHub

paidf-augmentation

Author and validate PAIDF augmentation YAML configs for remote Cosmos Transfer, Predict, image-edit, and image-to-video inference.

3.2k|370|Updated Feb 25, 2026
Search Tags:#docker#video generation#augmentation#yaml config#image edit#physical ai#cosmos transfer
One-click install
npx skills add https://github.com/NVIDIA/skills --skill paidf-augmentation
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: paidf-augmentation
Source: https://github.com/NVIDIA/skills/tree/main/skills/paidf-augmentation
Command: npx skills add https://github.com/NVIDIA/skills --skill paidf-augmentation

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

Configuring and running the PAIDF augmentation pipeline requires choosing the right generative model, writing schema-valid YAML configs, wiring captioning and evaluator endpoints, and launching a Docker container correctly — a process with many failure points that this Skill guides an agent through end to end.

Core Features & Use Cases

  • Model Selection: Picks the correct model from input type and goal — Cosmos Transfer 2.5 for video scene-attribute changes, Cosmos Predict 2.5 for text/image/video-to-video generation, image-edit for single-image edits, and Cosmos3 or Veo 3.1 for image-to-video.
  • Config Authoring & Validation: Writes and validates YAML configs against the PipelineConfig Pydantic schema, covering data, endpoints, captioning, augmentation, data_processing, and evaluators sections.
  • Captioning & Evaluators: Configures six captioning strategies (VLM+LLM, LLM-only, text, file, VLM-only, VLM+template) and quality evaluators such as hallucination checks and attribute verification.
  • Use Case: A robotics engineer wants to re-render a traffic video as a rainy night scene. The Skill selects Cosmos Transfer 2.5, authors a config with a video_transfer endpoint and VLM+LLM captioning, and provides the exact docker run and uv run commands to execute it.

Quick Start

Ask your agent to augment a video by changing its weather to a rainy night using the PAIDF augmentation pipeline and have it produce the config and launch commands.

Frequently Asked Questions about paidf-augmentation

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I choose between Cosmos Transfer and Cosmos Predict for video augmentation?▼

Use Cosmos Transfer 2.5 when you have an input video and want to change scene attributes like weather or lighting while preserving structure and motion. Use Cosmos Predict 2.5 when generating new video from text, image, or video conditioning, such as text2world generation with no media input.

How do I configure captioning for image attribute augmentation?▼

Use LLM-only captioning by setting captioning.llm with a system_prompt and variables mapping attributes to value lists, plus an llm-role endpoint in the endpoints list. A VLM section is not required for image-edit flows, and combining vlm with llm.text or llm.file_path is invalid.

Does the PAIDF augmentation pipeline require a GPU?▼

Remote inference needs no GPU because all models run behind remote HTTP endpoints with no local weights. A CUDA GPU is only required for the data_processing.alignment post-processor and for decoding H.264 video in evaluators or transcoding.

Why does my run fail with no endpoint matched selector cosmos-transfer2.5?▼

The model name must resolve to an entry in the endpoints list by id, then role, then the model-name-to-role map. The error means no endpoint with role video_transfer exists; add one with a url and model, where the adapter defaults to nim.

Can the PAIDF augmentation pipeline train or fine-tune models?▼

No, the pipeline performs inference and augmentation only and does not train or fine-tune models. It also does not deploy clusters or NIM endpoints; those tasks are explicitly out of scope.

How are API keys handled in augmentation configs?▼

Keys are never stored in YAML; each endpoint names an environment variable via api_key_env, such as VLM_API_KEY or VEO_API_KEY, and the value comes from the environment. Local endpoints like vLLM servers need no key at all.

Related Skills

Explore More Software Engineering Skills→
NVIDIANVIDIA
49

vss-call-vios-api

Manage cameras, streams, recordings, and storage via the VIOS API.

Official
Intermediate
Bakery88Bakery88

comfyui-lora-data-gathering

Gather, curate, and auto-caption LoRA training data into a ready dataset folder.

Community
Advanced
NVIDIA-AI-IOTNVIDIA-AI-IOT
110

jetson-llm-serve

Deploy OpenAI-compatible LLM or VLM serving endpoints on Jetson devices.

Official
Advanced
GarrettRoiGarrettRoi

segment-anything-model

Segment arbitrary objects in images using point, box, or mask prompts.

Community
Intermediate