Instructions to use nvidia/Cosmos3-Super-Text2Image with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Cosmos
How to use nvidia/Cosmos3-Super-Text2Image with Cosmos:
# No code snippets available yet for this library. # To use this model, check the repository files and the library's documentation. # Want to help? PRs adding snippets are welcome at: # https://github.com/huggingface/huggingface.js
- Diffusers
How to use nvidia/Cosmos3-Super-Text2Image with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("nvidia/Cosmos3-Super-Text2Image", dtype=torch.bfloat16, device_map="cuda") prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k" image = pipe(prompt).images[0] - Notebooks
- Google Colab
- Kaggle
| """Extract best agentic upsampling images from an output directory.""" | |
| from __future__ import annotations | |
| import argparse | |
| import csv | |
| import json | |
| import shutil | |
| from pathlib import Path | |
| from typing import Any | |
| from agentic_upsampling.io_utils import append_jsonl | |
| IMAGE_SUFFIXES = {".jpg", ".jpeg", ".png", ".webp"} | |
| def parse_args() -> argparse.Namespace: | |
| parser = argparse.ArgumentParser(description=__doc__) | |
| parser.add_argument("--output-dir", type=Path, required=True, help="Agentic upsampler run output directory.") | |
| parser.add_argument( | |
| "--export-dir", | |
| type=Path, | |
| default=None, | |
| help="Directory for copied best images and manifests. Defaults to OUTPUT_DIR/best_generations.", | |
| ) | |
| parser.add_argument("--overwrite", action="store_true", help="Replace existing copied images/manifests.") | |
| return parser.parse_args() | |
| def iter_best_jsons(output_dir: Path) -> list[Path]: | |
| """Return per-prompt best.json files in deterministic order.""" | |
| return sorted(path for path in output_dir.glob("*/best.json") if path.parent.name != "best_generations") | |
| def resolve_image_path(raw_path: str, *, output_dir: Path, best_json_path: Path) -> Path: | |
| """Resolve image paths written by runs launched with relative or absolute output dirs.""" | |
| image_path = Path(raw_path) | |
| candidates = [image_path] | |
| if not image_path.is_absolute(): | |
| candidates.extend( | |
| [ | |
| output_dir / image_path, | |
| output_dir.parent / image_path, | |
| best_json_path.parent / image_path.name, | |
| ] | |
| ) | |
| for candidate in candidates: | |
| if candidate.exists(): | |
| return candidate | |
| raise FileNotFoundError(f"Best image does not exist: {raw_path}") | |
| def copied_image_name(record: dict[str, Any], image_path: Path) -> str: | |
| """Build a simple copied image filename.""" | |
| prompt_id = str(record["prompt_id"]) | |
| suffix = image_path.suffix.lower() | |
| if suffix not in IMAGE_SUFFIXES: | |
| suffix = ".jpg" | |
| return f"{prompt_id}{suffix}" | |
| def extract_record(best_json_path: Path, *, output_dir: Path, images_dir: Path, overwrite: bool) -> dict[str, Any]: | |
| """Copy one best image and return its export manifest record.""" | |
| best_data = json.loads(best_json_path.read_text(encoding="utf-8")) | |
| if not isinstance(best_data, dict): | |
| raise ValueError(f"{best_json_path} must contain a JSON object.") | |
| best = best_data.get("best") | |
| if not isinstance(best, dict): | |
| raise ValueError(f"{best_json_path} is missing best candidate metadata.") | |
| raw_image_path = str(best.get("image_path") or "") | |
| if not raw_image_path: | |
| raise ValueError(f"{best_json_path} best candidate is missing image_path.") | |
| image_path = resolve_image_path(raw_image_path, output_dir=output_dir, best_json_path=best_json_path) | |
| record = { | |
| "prompt_id": str(best_data["prompt_id"]), | |
| "prompt": str(best_data.get("prompt") or ""), | |
| "best_score": best_data.get("best_score"), | |
| "best_iteration": best_data.get("best_iteration"), | |
| "selected_sample_index": best.get("selected_sample_index", best.get("sample_index")), | |
| "threshold_cleared_any": bool(best_data.get("threshold_cleared_any")), | |
| "source_image_path": str(image_path), | |
| "best_json_path": str(best_json_path), | |
| "analysis_path": str(best.get("analysis_path") or ""), | |
| } | |
| dest_path = images_dir / copied_image_name(record, image_path) | |
| if dest_path.exists() and not overwrite: | |
| raise FileExistsError(f"Refusing to overwrite existing image: {dest_path}") | |
| images_dir.mkdir(parents=True, exist_ok=True) | |
| shutil.copy2(image_path, dest_path) | |
| record["copied_image_path"] = str(dest_path) | |
| return record | |
| def write_csv(path: Path, records: list[dict[str, Any]]) -> None: | |
| """Write a flat CSV summary for quick spreadsheet inspection.""" | |
| fieldnames = [ | |
| "prompt_id", | |
| "best_score", | |
| "best_iteration", | |
| "selected_sample_index", | |
| "threshold_cleared_any", | |
| "copied_image_path", | |
| "source_image_path", | |
| "best_json_path", | |
| "analysis_path", | |
| "prompt", | |
| ] | |
| with path.open("w", newline="", encoding="utf-8") as f: | |
| writer = csv.DictWriter(f, fieldnames=fieldnames) | |
| writer.writeheader() | |
| for record in records: | |
| writer.writerow({key: record.get(key, "") for key in fieldnames}) | |
| def extract_best_images(output_dir: Path, export_dir: Path, *, overwrite: bool = False) -> list[dict[str, Any]]: | |
| """Copy best images from a run and write JSONL/CSV manifests.""" | |
| output_dir = output_dir.expanduser() | |
| export_dir = export_dir.expanduser() | |
| if not output_dir.exists(): | |
| raise FileNotFoundError(f"Missing output directory: {output_dir}") | |
| best_jsons = iter_best_jsons(output_dir) | |
| if not best_jsons: | |
| raise RuntimeError(f"No per-prompt best.json files found under {output_dir}") | |
| images_dir = export_dir / "images" | |
| manifest_path = export_dir / "best_generations.jsonl" | |
| csv_path = export_dir / "best_generations.csv" | |
| if overwrite: | |
| manifest_path.unlink(missing_ok=True) | |
| csv_path.unlink(missing_ok=True) | |
| elif manifest_path.exists() or csv_path.exists(): | |
| raise FileExistsError(f"Export manifests already exist in {export_dir}; pass --overwrite to replace them.") | |
| records: list[dict[str, Any]] = [] | |
| for best_json_path in best_jsons: | |
| record = extract_record(best_json_path, output_dir=output_dir, images_dir=images_dir, overwrite=overwrite) | |
| records.append(record) | |
| append_jsonl(manifest_path, record) | |
| write_csv(csv_path, records) | |
| return records | |
| def main() -> int: | |
| args = parse_args() | |
| export_dir = args.export_dir or (args.output_dir / "best_generations") | |
| records = extract_best_images(args.output_dir, export_dir, overwrite=args.overwrite) | |
| print(f"Exported {len(records)} best images to {export_dir}", flush=True) | |
| print(f"Images: {export_dir / 'images'}", flush=True) | |
| print(f"JSONL: {export_dir / 'best_generations.jsonl'}", flush=True) | |
| print(f"CSV: {export_dir / 'best_generations.csv'}", flush=True) | |
| return 0 | |
| if __name__ == "__main__": | |
| raise SystemExit(main()) | |