Skip to content

Tokenhub Update: MiniMax H3-Image Image Generation Models Released โ€‹

Updated

Tokenhub image generation adds two MiniMax H3-Image models (released 2026-11-18). Calls are synchronous: a single request returns the result directly in the response body, with no task polling.

ModelmodelHighlights
MiniMax H3-Imageminimax-image-h3Standard quality; text-to-image and multi-image reference editing with balanced results
MiniMax H3-Image Highspeedminimax-image-h3-highspeedHigh-speed version for latency-sensitive scenarios

1. Endpoints โ€‹

  • Request domain (international site): tokenhub-intl.tencentcloudmaas.com (Singapore)
  • Text-to-image: POST /v1/wand/minimax-image/generation
  • Image editing: POST /v1/wand/minimax-image/edits (1โ€“16 reference images)
  • Authentication: Authorization: Bearer <API_KEY>

2. Key capabilities โ€‹

  • Images per request n: 1โ€“10.
  • Pick one of three ways to control size: size (exact width ร— height) / aspect_ratio + area (1K / 2K / 4K, default 4K) / none (the model decides). Passing aspect_ratio / area together with explicit size dimensions returns 400.
  • Output formats: jpeg (default) / png / webp; return type: url / b64_json.
  • aigc_watermark adds a visible AIGC watermark; disable_prompt_enhanced turns off Prompt enhancement.
  • Before use, enable pay-as-you-go for the model in the console under Online Inference - Vision Models.

3. Docs โ€‹

Built with VitePress ยท Deployed on Tencent EdgeOne Pages