What's Changed

• Feat/router by @frontboat in #1

New Contributors

• @frontboat made their first contribution in #1

Full Changelog: v0.1.9...v0.1.10

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog,
and this project adheres to Semantic Versioning.

[0.1.9] - 2025-09-12

Added

• Comprehensive test suite with pytest
• Auto-generated API documentation with mkdocs
• LLM-friendly documentation (llm.txt) auto-generation
• Makefile for common development tasks
• Complete Google-style docstrings for all modules

Changed

• Reorganized package structure to use src/ layout
• Improved imports and module organization

Fixed

• Various import issues and path handling

[0.1.8] - 2025-08-17

Added

• Caching features for improved performance
• Safety settings support for generation
• Cached content support

Changed

• Reorganized package structure to use src/ layout
• Moved API-related modules to api/ subdirectory
• Moved MCP server to server/ subdirectory

[0.1.7] - Previous Release

Added

• Initial release features

veotools 0.1.8 — Release Notes

• Version bump

  • Package: 0.1.8
  • google-genai: upgraded to 1.30.0 (requirements and pyproject)

• SDK: generation config and validation

  • Safety settings: pass-through to video generation
    • Accepts list of dicts {category, threshold} or types.SafetySetting
    • Safe fallback: drops unknown fields on older clients
  • Cached content: pass-through via cached_content in config
  • Person generation rules enforced (fail-fast ValueError if invalid)
    • Veo 3: text→video = allow_all; image/video-seeded = allow_adult
    • Veo 2: text→video = allow_all|allow_adult|dont_allow; image/video-seeded = allow_adult|dont_allow

• CLI

  • New flags:
    • --safety-json to pass safety settings (JSON list)
    • --cached-content to reference existing cached content
  • Supported in both generate and continue

• MCP server and APIs

  • New explicit caching tools:
    • cache_create_from_files(model, files, system_instruction?)
    • cache_get(name)
    • cache_list()
    • cache_update(name, ttl_seconds?, expire_time_iso?)
    • cache_delete(name)
  • Existing tools unchanged: preflight, version, list_models, generate_start, generate_get (with streamed progress), generate_cancel, continue_video
  • Resources: recent videos (videos://recent/{limit}), job records (job://{job_id})
  • Still stdio transport via FastMCP; uv/uvx-friendly entrypoints

• Entry points

  • Console scripts:
    • veo-mcp (existing)
    • mcp-veo3 (new alias)

• Documentation

  • README updated:
    • Safety settings and caching support
    • CLI usage with --safety-json and --cached-content
    • Person generation constraints documented
    • Cache management examples: list/get/update/delete
  • Notes on output directory (VEO_OUTPUT_DIR) and defaults

• Backward-compatibility notes

  • New validation may raise ValueError for invalid person_generation combos that previously passed-through
  • Safety/cached_content fields are best-effort: automatically removed if not recognized by the installed client

• Quality of life

  • genai==1.30.0 upgrade
  • Safety settings pass-through
  • Caching helpers and usage
  • MCP server aligned with uv best practices and expanded tools

Full Changelog: v0.1.7...v0.1.8

• Package and naming

  • Renamed import surface to veotools; published-friendly pyproject.toml with console scripts (veo, veo-mcp)
  • Clean runtime deps; optional extra veotools[mcp]
  • Version sync; added LICENSE and release version-check script

• SDK improvements

  • Generation: added aspect_ratio, negative_prompt, person_generation passthrough (validated AR per model)
  • Output metadata: precise via ffprobe with OpenCV fallback
  • Storage: defaults to current working directory ./output (env override VEO_OUTPUT_DIR)
  • Model discovery: list_models(include_remote) with on-disk cache (10 min)

• MCP surface

  • Job lifecycle: generate_start, generate_get(wait_ms), generate_cancel
  • High-level: continue_video(video_path, prompt, …) (generate-and-stitch, progress streaming)
  • Resources: videos://recent/{limit}, job://{job_id}
  • Built-in server entry point: veo-mcp (stdio), with detailed parameter docstrings

• CLI

  • veo preflight, veo list-models, veo generate, veo continue
  • Flags: --model, --aspect-ratio, --negative-prompt, --person-generation, --extract-at, --overlap, --json
  • Progress bar output; paths saved under ./output

• Examples and docs

  • Examples updated to use model discovery and current APIs
  • README tightened: features, SDK quick start, CLI usage, MCP usage, Cursor mcp.json examples
  • Removed noisy sections; clarified behavior (overlap trim, storage location)

• Cleanup

  • Deleted temporary veo_stitch.py demo
  • Added GitHub Actions workflow guidance for Trusted Publishing (OIDC)

Impact

• One-command MCP server and CLI
• Safer, more configurable generation
• Better defaults (local output), richer metadata, clear docs and UX

add cli

You can now run:
veo generate --prompt "cat riding a hat" --model veo-3.0-fast-generate-preview
veo continue --video dog.mp4 --prompt "the dog finds a treasure chest" --overlap 1.0
veo --help

Full Changelog: v0.1.5...v0.1.6

update package naming to be consistent

Full Changelog: v0.1.4...v0.1.5

maybe it works now

how does this work

fix ci again

update readme and ci