OPEN BANANA Fixed Main Body Creativity

This is our open-source banana, a small application capable of contextual reasoning. The suggested input image resolution should be at least 1920x1080. Larger input images can effectively improve reasoning quality. The reasoning speed is not particularly fast, and it supports output within 2K resolution with any aspect ratio. However, 1280x720 is always the most efficient resolution.

1) The greater the main weight, the greater the imagination, but if it's too large, it may cause the image to collapse or the character features to shift. Default value: 0.4

2) You can mount style lora or mount the contextual lora from the previous version. However, if the weight is too large, it may cause character features to shift. Default value: 0.3

3) Prompt: Advertising design. It can be used for designing advertisements.

4) Prompt: Shot/reverse shot. It can be used for reverse shot angles.

5) Prompt: Another angle. It can still fix the scene and look for other visual angles, a very practical feature. As long as you enter this prompt, the camera will not leave the current scene.

6) Prompt: Hand close up. Used to focus on hand close-ups.

7) Prompt: Imagine. It improves reasoning quality to some extent, but due to the limitations of the training sample size, the actual improvement is not as significant as expected!

8) Prompt: OSM style. It improves reasoning quality to some extent, but due to the limitations of the training sample size, the actual improvement is not as significant as expected!