To start a good image, either use a template or have a real photo as a reference next to mokker opened in another tab, e.g. from Pinterest. The more of “this kind of image” the AI probably has seen before in our training, the more likely it is to generate good results.
When generated images look bad, this is often due to a perspective mismatch between two or more props and products. While the AI can deal with some inaccuracies, the overall quality declines the further the perspectives are off.
Here is an example of a bad perspectives:
The left chair prop is clearly visible from above, while the photo of the couch is taken from the front. The AI has no way of solving this impossible task, resulting in an odd image.
Use props that match the perspective instead, like in the second examle, resulting in a much more believable image.