tongflow-modal-gemma4

June 17, 2026 · View on GitHub

Official TongFlow plugin. Multimodal understanding with Gemma-4 (google/gemma-4-E4B-it), running on a GPU via Modal. Describes or answers questions about images and video.

Capabilities

  • Image understanding (image-gen-text) — captions, Q&A, or descriptions from an image.
  • Video understanding (video-gen-text) — summaries or descriptions from a video.

Credentials

Add in TongFlow Settings (gear icon, top-right):

KeyRequiredNotes
MODAL_TOKEN_IDCreate at modal.com/settings/tokens.
MODAL_TOKEN_SECRETPaired with MODAL_TOKEN_ID.

On first use the plugin deploys to your Modal account automatically and caches the build. The google/gemma-4-E4B-it weights are public — no Hugging Face token required.