-
Notifications
You must be signed in to change notification settings - Fork 80
Description
What part of the documentation do you want to update?
I want to update the p-image-edit Performance Model documentation.
Currently, the documentation defines the API parameters (e.g., images, prompt, strength) but lacks a specific guide on how to structure the text prompt itself.[1] Users coming from Midjourney or DALL-E often fail because they use descriptive prompts ("A boy sitting on a chair") rather than the imperative/command-based prompts required by Pruna's editing architecture ("Place the boy from Image 1 on a chair").
What changes do you want to make?
I propose adding a "Prompt Engineering Best Practices" section to the p-image-edit page that outlines the following logic for better user results:
- The "Imperative" Rule (Instruction vs. Description)
Explain that p-image-edit acts like a photo editor, not a painter.
Bad: "A cute boy standing in a forest."[1]
Good: "Place the boy from Image 1 in a forest. Make him look at the camera."
- The Multi-Image Reference Syntax
Explicitly document the syntax for anchoring characters when using the images array:
"Using the character from Image 1, [action]..."
Explain that this specific phrase triggers better consistency than generic terms like "the boy" or "he."
-
The "Golden Formula" for Scene Composition
Suggest a standard prompt structure for consistent scenes:
[ACTION & PLACEMENT] + [EMOTION & POSE] + [BACKGROUND & COMPOSITION] -
Composition & Negative Space Tips
Add a specific tip for users creating layouts for text overlay (e.g., book covers, ads).
Explain that standard prompts center the subject, so users must be aggressive with directional commands:
"Position the character on the far right third of the image, leaving the left side as open negative space."
Sources
help
pruna.ai
Google Search Suggestions
Display of Search Suggestions is required when using Grounding with Google Search. Learn more
Pruna AI p-image-edit prompt syntax guide
Pruna p-image-edit documentation prompt engineering
Pruna p-image-edit imperative vs descriptive prompt