image-cog

**Judgement:** A strong, versatile image-generation skill well-suited for production-like workflows that need multi-model coverage (photoreal, transparent PNGs, vectors) and structured image tasks. Best used inside the CellCog ecosystem where its async patterns and agent routing are already expected. **What it’s good for:** - Marketing and product teams needing consistent, professional imagery (hero shots, social sets, blog art, ad variations). - Creators who need multi-image coherence: character series, storyboards, mascot sets, campaign themes. - Technical users who want a single chat-based entry point to Google, OpenAI, and Recraft image models with automatic routing. **Key strengths:** - Broad functional coverage: text-to-image, image editing, style transfer, reference-based generation, and cohesive image sets. - Smart model selection (Nano Banana 2 default, GPT Image 1.5 for transparent PNGs, Recraft for vectors) without extra wiring. - Clear guidance on prompt design, image specs (aspect ratios, resolutions), and when to use `agent` vs `agent team`. **Risks / limitations:** - Hard dependency on the separate `cellcog` skill for SDK setup; not a standalone image API wrapper. - Reliant on external providers (Google, OpenAI, Recraft), so behavior, quality, and quotas can change outside your control. - No explicit, documented custom safety or compliance layer beyond what underlying models likely enforce—be careful with sensitive or IP-heavy inputs (faces, brands, copyrighted styles). - Async “fire-and-forget” pattern may add integration complexity for very simple, synchronous workflows. **Recommended scenarios:** - You already use CellCog or are comfortable adding it and want a high-level, multi-model image layer with good documentation. - You need recurring, consistent visual assets (product lines, characters, brand imagery) rather than one-off experimental generations. - You want to offload model choice and routing logic to an agent layer instead of managing multiple image APIs directly.