When generating portraits or some basic landscapes, you will get some basic or even random camera angles. And you don’t really actively think about camera angles, but when you are generating your image for use somewhere, you want it to have some meaning, and camera angles are a well-known method of storytelling and compelling composition, and they can also help you convey the intention or intensity behind your image.
Most of you might know some basic camera angles like wide angle, midshot, and some others, but there are many dynamic camera angles that are often used in movies, mangas, comics, and other storytelling mediums.
In this post, we will show you my findings when trying to generate images with some dynamic camera angles.
Related: Best Stable Diffusion Models For Realism
What is camera angle
Camera angle refers to the position of the camera in relation to the subject being photographed. Different camera angles can convey different emotions, perspectives, and meanings. For instance, an over-the-shoulder shot can create a sense of intimacy, while a high-angle shot can create a feeling of vulnerability.
- Over the shoulder shot
- Hip Level Shot
- Shoulder Level Shot
- Low angle
- Point-of-View Shot
Ways to controlling camera angles
To control camera angle in stable diffusion, one needs to consider several factors.
The camera angle of the photo needs to be described.
Also, many times, describing an action or event with a camera angle can help you get the best possible results, and you have to understand that this works well for simple camera angles like low angle, high angle, and birds eye view.
camera looking up to young man standing in the road of dystopian scifi city, tilt view, low angle,
Some camera angles may only work with certain image aspect ratios, so it is important to try out different aspect ratios to see what works best.
Additionally, adding weight to the camera angle prompt may help produce better results, but it can also make it worse if it’s too much.
camera looking up to young man standing in the road of dystopian scifi city, tilt view, low angle, (low angle), (camera from foot looking up: 1.3), (grounded camera looking up)
Another way to increase the likelihood of generating images with a specific camera angle is to use multiple keywords related to that angle. For example, if you want to generate an image with a hip-level shot, using keywords such as “hip level,” “low angle,” and “ground-level” may help improve the results.
camera looking up to young man standing in the road of dystopian scifi city, tilt view, low angle, (low angle), (camera from foot looking up), grounded camera looking up
Sometimes, simplifying the desired camera angle can also be helpful. For instance, for a hip-level shot, using the prompt “hip high camera angle” may produce better results than a more complex description.
camera from foot looking up to young man standing in the road of dystopian scifi city, tilt view, low angle, low angle, camera from foot looking up
It is also important to consider negative prompts when controlling the camera angle. If the generated images are not aligned with the desired camera angle, adding the unwanted angle as a negative prompt may help.
ControlNet can be used to create images with your desired camera angle. This can be done through depth or, canny, or by providing an image of the desired camera angle and using ControlNet to see what produces the best results.
And this needs At least a decent GPU with 8GB of VRAM minimum if you looking for a budget GPU for stable diffusion you can read this article.
ControlNet can be used to define the scene through scribbles. This allows for the description of the composition, which includes camera angle. By providing a more detailed description of the scene, ControlNet can generate more accurate images.
If you know about blender then it will be easy for you to adjust camera angles, as you can quickly make base image in blender your camera angles then use that to drive your image generation.
- To control camera angle in stable diffusion, start by describing the camera angle of the photo and trying out different aspect ratios.
- Adding weight to the camera angle prompt and using many keywords related to the desired camera angle can increase the chances of generating images with that camera angle.
- Sometimes it may be necessary to oversimplify the desired camera angle, such as using “hip high camera angle” for a hip angle shot.
- If the generated image comes with an undesired camera angle, adding that angle in negative prompt can help.
- ControlNet can be used to give a general idea of camera angle using depth or canny, or by inputting an image of the desired camera angle.
- Scribbling in ControlNet to define the scene can also help describe the composition, including camera angle.
In conclusion, controlling camera angle in stable diffusion can be challenging, but with the use of ControlNet, it becomes easier. By describing the camera angle, using multiple keywords, simplifying descriptions, and using ControlNet, one can generate images with the desired camera angle. With ControlNet, there is no need to worry about destroying the production-ready diffusion model, making it ideal for use on small-scale or personal devices.