Skip to main content

​
1. OpenAI compatible format

Recommended for most new integration scenarios:
  • POST /v1/chat/completions
  • POST /v1/responses
  • GET /v1/responses/{response_id}
  • POST /v1/images/generations
  • POST /v1/images/edits
  • POST /v1/videos
  • GET /v1/videos/{task_id}
Features:
  • SDK ecosystem is mature
  • Consistent field styling
  • Ideal for switching between multiple models

​
2. Model Native Format

Expose native interfaces when the model requires dedicated fields, specific task structures, or native query protocols:
  • Seedance：/api/v3/contents/generations/tasks
  • HappyHorse: /api/v1/services/aigc/video-generation/video-synthesis and /api/v1/tasks/{task_id}
  • Vidu：/ent/v2/text2video、/ent/v2/img2video、/ent/v2/start-end2video、/ent/v2/reference2video 和 /ent/v2/tasks/{task_id}/creations
  • Omni and Veo: gemini-omni uses /v1/videos or /api/v3/contents/generations/tasks; the native reserved paths for Veo are /v1beta/models/{model}:predictLongRunning and /v1beta/{operation_name}
  • Gemini Image Generation: /v1beta/models/{model}:generateContent
  • Music generation: /suno/submit/music and /suno/fetch/{task_id}, supports prompt description mode, light music/instrumental mode, and advanced lyrics mode
gemini-omni is the publicly exposed video model name; you don’t need to worry about the internal upstream model. For video calls, use /v1/videos first; you can also use the Seedance task format /api/v3/contents/generations/tasks. mode=t2v/r2v/edit represent text-to-video, reference-based generation, and video editing respectively, and the duration field will be automatically mapped to the 4 / 6 / 8 / 10-second tiers. Best for:
  • Large number of model-specific fields
  • Requires native request body structure
  • Needs to remain close to the upstream official format

​
3. Model-Specific Video Interface

A few model-specific interfaces are currently available, such as:
  • POST /kling/v1/videos/text2video
  • POST /kling/v1/videos/image2video Best for:
  • Existing integrations based on the specific model protocol
  • Requires specific parameter modes for that model

​
4. Selection Recommendations

  1. For new projects, prioritize the OpenAI compatible format.
  2. Switch to the native format only if the compatible format cannot cover model-specific fields.
  3. Long-running task interfaces should clearly distinguish between “Create API” and “Query API,” and save task_id or operation_name.
  4. When multiple invocation formats exist for the same model, refer to the supported fields on the corresponding API page; do not mix request body formats.