ComfyUI, image APIs chase storybook consistency

Hot take: this is mostly a state-management problem, not an image-model problem.

• –You can keep generation serverless, but character consistency still needs durable external state: a character bible, reference images, fixed seeds, or stored style tokens.
• –ComfyUI now has cloud/API options, but the tradeoff is still workflow complexity; it is stronger for graph-heavy control than for a lightweight app backend.
• –OpenAI's image API and Responses API can work with image context, but the docs still warn that recurring characters can drift between generations.
• –For kids' storybooks, the cleanest setup is local LLM for scene planning, remote image API for rendering, and a thin persistence layer for character assets and prompts.