Reve 2.0 AI

Reve 2.0 AI is a generative media platform that exposes both image and video generation through a single REST API endpoint. According to the official documentation, the platform supports four distinct image models—Reve 2.0, Nano Banana, Nano Banana Pro, and Seedream 5—alongside a roster of scene-specific video models that handle text-to-video, image-to-video, and video-to-video workflows.

How It Works

All generation follows the same two-step pattern. A client sends a POST request to /api/v1/generate specifying a model identifier, a text prompt, and optional parameters such as aspect ratio, duration, resolution, or reference media URLs. The server returns a task_id. The client then polls GET /api/v1/status?task_id=... until the task resolves to SUCCESS or FAILED.

The vendor states that credits are deducted at task creation, with the consumed amount returned in the success response. Pricing depends on the model, resolution, duration, and any additional options selected. Failed tasks are refunded automatically. API keys are created and managed through the Settings → API Keys dashboard and must be kept server-side; the documentation explicitly warns against embedding keys in client-side code.

Image Generation

The image pipeline supports both text-to-image and image-to-image editing modes. Supplying image_urls in the request payload switches a model into edit mode. The flagship Reve 2.0 model accepts parameters including aspect_ratio, test_time_scaling, and an upscale toggle. Nano Banana Pro is positioned for editing tasks—the documentation shows an example of modifying an uploaded photo's sky to a dramatic sunset. This places Reve 2.0 AI alongside other tools in the AI Image Recognition space, though its emphasis is on generation and editing rather than classification.

Video Generation

Video models are scene-specific, meaning each model variant targets a particular input modality. The documented lineup includes happyhorse-1.0-text-to-video for text-driven clips, kling3-pro-image-to-video for animating still frames with camera-motion prompts, and kling26-motion-control-720p for video-to-video motion transfer. Videos support configurable duration—the examples use 5 seconds—and resolution up to 720p.

Developer Experience

The unified API surface is the platform's primary architectural decision. Rather than separate endpoints per modality, every generation request—image or video, creation or edit—goes through the same /api/v1/generate path with model selection handled by a field in the JSON body. This design simplifies integration for developers who need both image and video output. Compared to purpose-built alternatives like Describe Picture&Image, Reve 2.0 AI trades specialized depth for breadth across modalities.

Practical Considerations

The credit-based billing model means costs are transparent at task creation—you know the deduction before the generation completes. Automatic refunds on failure reduce the financial risk of long-running or experimental prompts. However, the documentation does not publish a complete pricing table inline; users must consult per-category pages to estimate costs for specific model, resolution, and duration combinations. The platform's reliance on a single async polling pattern also means real-time streaming or progressive output is not part of the documented workflow. Developers evaluating the platform against alternatives like Molmo AI should weigh the breadth of modalities against the absence of synchronous or streaming generation paths.