Skip to main content

New model references and capability updates

Wan 3.0 Video Standard

Added wan3.0-video for text, first/last frames, and multimodal references. See Wan 3.0 Video and pricing for current capabilities and rates.

Unified Video Material Fields

Image, video and audio inputs in video-generation requests now use reference_type, defaulting to asset. Editing requires an explicit reference_type: edit_source. Frame roles remain first_frame and last_frame.The old refer_type field no longer selects the material’s purpose. Omitted, null, empty or unknown types are treated as asset. Update your requests, especially those relying on the old base default to trigger editing. keep_original_sound remains a per-video property. Consult each model’s API page for supported inputs and combinations.Credit unit prices are unchanged.

New video model: HappyHorse 1.0

Released happyhorse-1.0 — a single model ID that covers text-to-video, image-to-video (first frame), reference-to-video (multi-image), and video editing (video + reference images), auto-dispatched by input shape.
  • Supports 720P / 1080P output, duration 3s–15s (video-edit mode follows the source video length, capped at 15s);
  • In reference-to-video mode, refer to the N-th image with character1, character2, … in the prompt;
  • Video-edit mode exposes extra_params.audio_setting (auto / origin) for audio handling;
  • Pricing: 720P 180 credits/second, 1080P 300 credits/second.
For full parameter details, see HappyHorse 1.0. For credit consumption, see Pricing.

First batch of Open Platform models now live

imini Open Platform is officially live, with the first wave of image and video generation models available.

Image models

  • nano-banana: Based on Gemini 2.5 Flash Image. Fast and low-cost, supports up to 3 reference images.
  • nano-banana-pro: Based on Gemini 3 Pro Image. Supports up to 14 reference images, output up to 4K.
  • nano-banana-2: Based on Gemini 3.1 Flash Image. Supports up to 14 reference images, with 512 / 1K / 2K / 4K output.

Video models

  • kling-v3: Kling 3.0 video generation. Supports 720P / 1080P with optional audio.
  • kling-v3-omni: Kling 3.0 full-modality variant. Supports reference video input.
  • seedance2.0: Doubao Seedance 2.0. Supports text-to-video, image-to-video, and multimodal references (image + video + audio).
  • seedance2.0-fast: Seedance 2.0 Fast — the same capabilities at a lower cost.
For detailed parameter documentation, see the API Reference. For credit consumption, see Pricing.