Yollomi AI Image Generator (All Models)

**Quick judgement:** A capable, feature-rich multi-model media generation skill that cleanly exposes the Yollomi image/video API through two tools (`yollomi.generate`, `yollomi.listModels`). Strong choice when you explicitly want to tap into a broad set of third-party image and video models via one interface, but it inherits all the usual external-API and safety limitations. **What it’s good for** - Centralizing access to many SOTA image/video models (Flux, Imagen, SD, Veo, Sora, Runway, etc.) behind one tool. - Building assistants that can both create and edit images (background removal, object remover, face swap, upscaling, restoration, virtual try-on). - Cost-aware agents that can query `listModels` to pick models based on credit price and capabilities. - Workflows needing configurable aspect ratios and multiple outputs from a single prompt. **Key risks and limitations** - **External dependency & credits:** All functionality depends on Yollomi’s availability, rate limits, and credit balance; 401/402 errors must be handled by the caller. - **Latency and throughput:** Actual performance is determined by Yollomi and the underlying model providers; may be slow for heavier models (e.g., high-end video or "pro" image models). - **Safety & content controls:** No explicit mention of built-in moderation or safety filters. If you need strict content controls (NSFW, violence, deepfakes, etc.), you must add your own policy checks and guardrails around this skill. - **Model-specific parameters:** Different models require different arguments (e.g., `imageUrl`, `mask`, `clothImage`, `personImage`, `inputs` for video). Incorrect parameterization can lead to failures unless the calling agent is careful or uses `listModels`/docs. **Recommended scenarios** Use this skill when you: - Want a single, unified way to call many image and video generation/editing models. - Are comfortable relying on an external paid API (with keys and credits managed out-of-band). - Can layer your own safety, validation, and error-handling logic around the tool. Avoid it (or wrap it heavily) in contexts where you need guaranteed offline operation, strict content moderation, or fully predictable latency.