Group Chat Plus attaches the StepFun image tools to the final reply phase instead of ordinary message preprocessing. The model first prepares a sendable image prompt, then calls text-to-image or image-edit tools; editing reads the first image from the current message so older context does not accidentally become the source material.
The implementation reuses API key, base URL, timeout, and proxy settings from the configured provider, while still allowing plugin-level image parameter overrides. The tool sends images internally and returns an empty result to the Agent, reducing the chance of the same image being sent again during response assembly. Tests cover tool registration, group permission checks, progress hints, stale capability-text filtering, and duplicate image handling.