Absolutely, preserving the composition and visual continuity when using transformative models like Sora can be challenging. Here are some tips to help minimize the “scene overwrite” or “page flipping” effect and keep your core structure intact:
1. Emphasize Core Elements in Your Prompt
Clearly state the elements you want to remain unchanged. Use phrases like:
- “Retain the original composition and pose.”
- “Keep the character’s silhouette and background layout identical.”
- “Do not alter the main subject’s position or orientation.”
You can also mention things you don’t want changed (negative prompting) explicitly, e.g., “Do not change the main character’s pose or the arrangement of the valley and encampment.”
2. Use an Image as a Strong Reference
If Sora allows for reference image uploads, make sure to select the input image as a base and set the transformation (or prompt guidance) strength to a lower value. This should tell the model you want more subtle, style-based transformation rather than a total reimagining.
3. Adjust Guidance/Prompt Strength
You’re already trying low prompt strength, but sometimes it may work better to experiment within an even smaller range or use “preserve composition/pose” slider settings if available.
4. Chained/Iterative Transforms
If a single transformation is too extreme, try applying smaller stepwise changes—run the model multiple times, gradually evolving style or details while checking for continuity.
5. Add Technical Descriptors
In your prompt, specify:
- “Subject and background to remain fixed.”
- “No changes to the character’s pose, position, or overall scene composition.”
- If possible, mention “do not flip, mirror, or relocate subjects.”
6. Negative Keywords
Include negative prompt terms like: “No new characters,” “No landscape changes,” “No changes to camera angle.”
7. Image Control Features
If Sora provides advanced options like segmentation masks, pose transfer, or depth maps, use them to lock key scene elements.
8. Provide a Style Reference, Not a Scene Reference
If your goal is to apply a visual style but keep the composition, specify: “Apply [style] while maintaining all layout and pose features of the original scene.”
Summary Sample Prompt
“Transform this image only by changing lighting and color grade—preserve the exact character pose, background structure, and arrangement of all scene elements. No new characters, no changes in perspective, and no scene shifting. The core composition, including character location, not to be altered.”
Model-Specific Advice
Every model differs in how strictly it follows image conditioning—if Sora supports ControlNet or similar image structure guidance, enable these. Purpose-built ‘structure-preserving’ models or tools may give better results.
If you’d like, share your exact Sora settings or some outputs for more tailored advice! Good luck—epic dark fantasy is a great testbed for these workflows.