Repository navigation
MeiGen 3.1.0 — Video References and Saved Characters
MeiGen 3.1.0 adds video/audio reference preparation and explicit content-image inputs for video generation. Matching remote services are deployed; npm meigen@3.1.0 is published as latest.
What is new
- Prepare selected MP4/MOV or WAV/MP3 references with
upload_reference_media. The returned public URLs are temporary, with seven-day cleanup. Preparation does not charge generation credits; purchased-balance, format, duration and account limits apply. - Use
generate_video.contentReferenceImagesseparately from first/last frames. Model discovery supplies supported combinations, reference syntax and the applicable pricing tier. - Read and reuse saved MeiGen characters through remote OAuth with the additional
meigen:characters:readpermission. Character creation is not included. Local npm and API keys cannot read private characters or libraries. - Preserve existing API-key generation, public search and Skills, with bounded local uploads and recovery using the original request ID and inputs.
Upgrade
New connections should use remote OAuth. Refresh or reconnect the host to discover new tools, then complete any requested additional consent. Remote MCP and the official ChatGPT adapter expose 22 tools; local npm exposes 17.
For an existing local integration, update only the package version in its saved command arguments to meigen@3.1.0; preserve credentials and other settings. Local image, video and audio inputs share a 256 MiB per-request limit.
Companion releases: ClawHub plugin meigen-ai-design@4.1.0 and standalone Skill creative-toolkit@3.2.0. Install only one copy of the Skill. Updated documentation.
HeyGen Video 1 was enabled for Web, API and MCP on October 12, 2026. It remains unavailable in native iOS/Android apps, and reference-video inputs remain disabled. Read the live model catalog for supported settings and rates.
The Git tag identifies the source used for the npm artifact. Registry package SHA-1: eab19f529e88188fe6f3dcd3a9993b5b041e2ce7. The packed server passed an actual SDK tool-discovery and public-tool-call check; the OSS suite passed 240 tests. A subsequent production check through the official ChatGPT adapter completed a HeyGen text-to-video request (480p, 5 seconds), saved its result and confirmed a single 10-credit purchased-balance charge. New authenticated media-import and character-consent flows still require separate host acceptance.