/generate-image

Generate or edit an image based on the user's request.

Process

1. Parse the arguments:

   • The main text is the image prompt
   • If --edit is present, the next argument is the path to an input image for editing
   • Determine mode: "edit" if --edit flag is present, otherwise "generate"

2. Ask the user which provider to use with AskUserQuestion:

   • Gemini (gemini-2.5-flash-image): Best for aspect ratio control and iterative editing
   • OpenAI (gpt-image-1.5): Best for text rendering and transparent backgrounds
   • xAI (grok-2-image): Flat per-image pricing, prompt revision, diverse styles
   • All in parallel: Generate with all available providers and pick the best result

3. Ask the user where to save the output with AskUserQuestion:

   • Current directory (e.g., ./generated-image.png)
   • Custom path (let them type)

4. Execute the generation using tasks for progress tracking:

   If single provider selected: a. Create a task with TaskCreate:

   • subject: "Generate image with <Provider>"
   • activeForm: "Generating image with <Provider>..." b. Mark it in_progress with TaskUpdate c. Run the script via Bash:

   bash "${CLAUDE_PLUGIN_ROOT}/scripts/<provider>.sh" \
     --mode <generate|edit> \
     --prompt "<prompt>" \
     --output "<output-path>" \
     [--input-image "<input-path>"]
   

   d. Mark the task completed (or note the error if it failed)

   If multiple providers selected: a. Create a task per provider with TaskCreate:

   • "Generate image with Gemini" (activeForm: "Generating image with Gemini...")
   • "Generate image with OpenAI" (activeForm: "Generating image with OpenAI...")
   • "Generate image with xAI" (activeForm: "Generating image with xAI...") b. Mark all in_progress with TaskUpdate c. Open a streaming display pane (single Bash call, capture the watch directory path):

   source "${CLAUDE_PLUGIN_ROOT}/scripts/display.sh" && display_pane_open
   

   This outputs a path like /tmp/display_pane.XXXXXX — capture it as DISPLAY_PANE_DIR. d. Launch Task subagents in parallel (subagent_type: Bash), each running one script. Use different output filenames (e.g., image-gemini.png, image-openai.png, image-xai.png). Pass DISPLAY_PANE_DIR so images appear progressively in the shared pane:

   DISPLAY_PANE_DIR=<captured-path> bash "${CLAUDE_PLUGIN_ROOT}/scripts/<provider>.sh" \
     --mode <generate|edit> --prompt "<prompt>" --output "<output-path>"
   

   e. As each subagent returns, mark its corresponding task completed (or note the error) f. After all providers complete, close the streaming pane to show interactive controls:

   DISPLAY_PANE_DIR=<captured-path> bash -c \
     'source "${CLAUDE_PLUGIN_ROOT}/scripts/display.sh" && display_pane_close'
   

5. After generation completes, confirm the output path(s) to the user. If multiple were generated, let the user know all files are available so they can compare.

Environment Requirements

• GEMINI_API_KEY for Gemini provider
• OPENAI_API_KEY for OpenAI provider
• XAI_API_KEY or GROK_API_KEY for xAI provider