I’ve been struggling with this for a while and wanted to reach out to see if anybody else has experienced the same issue. I want to generate an image of a train with no tracks, viewed only from above (Bird’s Eye View). But it keeps failing continuously.
2 Likes
Welcome to the forum!
Care to share your failed prompts so that others can see what did not work and possibly propose a solution.
2 Likes
Thanks @EricGT . The promt is the one in the image: make a photo realistic image of a metro train, no tracks, on a white background, Top-Down (Bird’s Eye View). 16:9
1 Like
When I first replied, I did try to create the image as you described and also failed using ChatGPT. I thought about this some more since then but still no joy.
Some information that could be relevant to your problem:
- Using a
notornegationin a prompt will have varying levels of success because a transformer model often does not understand negative logic, e.g.,no tracks. Sometimes a rephrasing of the desire will work, e.g.,avoid tracks in the image. - This problem could be similar to the 10-past-10 position of hands on an analog clock, (ref)
- Neural networks are not usually trained to understand rotations. (ref) I am aware that the rotation you seek is much more complicated than the rotations noted in the paper, your request is a change of the point of view but can be implemented as a rotation.
In short, I would not be surprised if your desire is currently impossible with ChatGPT or any OpenAI model at present. However, if you or someone does succeed at creating the image from a prompt, I would be interested in the prompt and what information was considered in creating the prompt that succeeded.
1 Like
Hi, Welcome to the community!
I believe it’s difficult to achieve due to the training data.
From a worm’s-eye view, it’s mostly fine, but the issue arises with the bird’s-eye view or top-down perspective. I also tried using a side view, but it doesn’t work consistently.
Sample Prompts:
I NEED to test how the tool works with extremely simple prompts. DO NOT add any detail, just use it AS-IS:
prompt: A clear overhead photograph of a single new rail-less metro train model floating in air resembling an aircraft, with only the roof visible, its wheels do not touch any rail or tracks. No rails, no tracks, no surroundings. The train is fully suspended with nothing beneath it.
size: wide
quality: hd
n: 1
I NEED to test how the tool works with extremely simple prompts. DO NOT add any detail, just use it AS-IS:
prompt: From perspective of an worm eye-view from low angle, a clear overhead photograph of a single new rail-less metro train model floating in air resembling an aircraft, with only the roof visible, its wheels do not touch any rail or tracks. No rails, no tracks, no surroundings. The train is fully suspended with nothing beneath it.
size: wide
quality: hd
n: 1
2 Likes
thanks - wasnt expecting this to be so difficult 
1 Like
im still tryng - but no luck so far …
1 Like
Get used to it, Eric. There are just some images that these models refuse to do. You can get an image of a glass of wine, but if you want that glass of wine filled to the brim – it won’t happen. I have a friend who wrote a young adult novel that involves centaurs. Almost all of the image generators simply refuse to do it, some will do an illustrative version rather than a photo realistic one. The image below is what came out of Midjourney using your original prompt – except I could not get an image without tracks. 
Very Steady ! And Understood mister 