silent oak
Summary
The image generation feature fails to execute prompts requesting the transformation of an illustration into a photorealistic image. Despite using progressively detailed and explicit prompts to create a photograph, the model consistently ignores the core request and outputs a near-duplicate of the original uploaded illustration. The model appears to be incapable of style transformation in this context, getting stuck on the source image's artistic style.
Detailed Steps to Reproduce
- Upload the attached image, which is a historical illustration.
Use the image generation feature with the initial prompt: Please transform this image into a realistic photograph, capturing the scene with photographic detail and depth, as if taken with a modern camera.
Expected Result: A new image that looks like a real photograph of the scene.
Actual Result: The original illustration is regenerated with almost no changes.
- Attempt again with a more forceful, detailed prompt:
Transform this image into a photorealistic, high-resolution photograph. Focus on rendering all elements with lifelike textures, natural lighting, and accurate shadows.
Actual Result: The model again returns the original illustration, ignoring the specific keywords like "photorealistic," "lifelike textures," and "natural lighting."
- Make a final, hyper-specific attempt:
Render this image as a highly realistic photograph, showcasing ultra-fine details and textures
Actual Result: The output is, once again, the same illustration. The core instruction to change the medium from illustration to photograph is completely disregarded every time.
Relevant Environment Details & Visual Evidence:
Platform: Gemini Web Application
Browser: Chrome
Device Type: Desktop
OS: Windows
Visual Evidence: The attached image serves as the Input. The Output in all test cases is a nearly identical version of this same file, which is the entire basis of this bug report.
robust flame