latentblending/parameters.md

46 lines
2.1 KiB
Markdown
Raw Permalink Normal View History

2023-01-09 12:47:07 +00:00
# Gradio Parameters
2023-01-09 12:42:02 +00:00
## depth_strength
2023-01-09 12:47:07 +00:00
- Determines when the blending process will begin in terms of diffusion steps.
- A value close to zero results in more creative and intricate outcomes, but may also introduce additional objects and motion.
- A value closer to one indicates a simpler alpha blending.
2023-01-09 12:42:02 +00:00
2023-02-20 10:26:41 +00:00
## branch1_crossfeed_power
- Controls the level of cross-feeding between the first and last image branch. This allows to preserve structures from the first image.
- A value of 0.0 disables crossfeeding.
- A value of 1.0 fully copies the latents from the first branch to the last.
## branch1_crossfeed_range
- Sets the duration of active crossfeed during development. High values enforce strong structural similarity.
- The value x ranges from [0,1], and the crossfeeding is deactivated after x*num_inference_steps steps
## branch1_crossfeed_decay
2023-02-20 10:44:50 +00:00
- Sets decay for branch1_crossfeed_power. Lower values make the decay stronger across the range.
2023-02-20 10:26:41 +00:00
- The value x ranges from [0,1], and the branch1_crossfeed_power is decreased until the end of the branch1_crossfeed_range to a value of x*branch1_crossfeed_power
## parental_crossfeed_power
Similar to branch1_crossfeed_power, however applied for the images withinin the transition.
## parental_crossfeed_range
Similar to branch1_crossfeed_range, however applied for the images withinin the transition.
## parental_crossfeed_power_decay
Similar to branch1_crossfeed_decay, however applied for the images withinin the transition.
2023-01-11 10:46:41 +00:00
2023-01-09 12:42:02 +00:00
## guidance_scale
2023-01-09 12:47:07 +00:00
- Higher guidance scale encourages the creation of images that are closely aligned with the text.
- Lower values are recommended for the best results in latent blending.
2023-01-09 12:42:02 +00:00
## guidance_scale_mid_damper
2023-01-09 12:47:07 +00:00
- Decreases the guidance scale in the middle of a transition.
- A value of 1 maintains a constant guidance scale.
- A value of 0 decreases the guidance scale to 1 at the midpoint of the transition.
2023-01-09 12:42:02 +00:00
## num_inference_steps
2023-01-09 12:47:07 +00:00
- Determines the quality of the results.
- Higher values improve the outcome, but also require more computation time.
2023-01-09 12:42:02 +00:00
## nmb_trans_images
2023-01-09 12:47:07 +00:00
- Final number of images computed in the last branch of the tree.
- Higher values give better results but require more computation time.