I’m building a system prompt through trial and error. The issue I’m currently stuck on is that I want to upload an image and edit the uploaded image, but it’s not working properly. When I edit an image generated by the action, it works fine. Is this perhaps a feature that’s fundamentally impossible? Are there any tips for the prompt? If anyone knows, I’d appreciate your input.
You’re not imagining it. Editing uploaded images is still a bit unpredictable across the board, even on native Google Gemini itself 
….this isn’t a Pickaxe-only limitation.
What you’re seeing makes sense:
-
Images generated by an Action are easier for the model to edit because it has full internal context.
-
User-uploaded images come in as external inputs, and models are less consistent about preserving or modifying them exactly as asked.
For Pickaxe Actions, it really does come down to prompting. A few things that usually help:
-
Be very explicit that the model must edit the uploaded image, not regenerate a new one.
-
Describe what must stay the same before describing what should change.
-
Keep edits small and specific. Big conceptual changes tend to break image continuity.
-
If possible, reference the image as “the uploaded image” consistently in the prompt.
So no, it’s not fundamentally impossible. It’s just one of those areas where current image models are still a bit fragile, and tight prompting makes a real difference. You’re on the right track experimenting.
2 Likes
Thank you so much for everything. Your explanations are very clear and easy to understand. When I explicitly framed the attached image as context for the AI to understand, just as you suggested, it worked perfectly!
2 Likes