better docs for regions
This commit is contained in:
parent
918ffde8e6
commit
30a9d01432
|
@ -34,6 +34,7 @@ Please see [the server admin guide](server-admin.md) for details on how to confi
|
||||||
- [Textual Inversion tokens](#textual-inversion-tokens)
|
- [Textual Inversion tokens](#textual-inversion-tokens)
|
||||||
- [Region tokens](#region-tokens)
|
- [Region tokens](#region-tokens)
|
||||||
- [CLIP skip tokens](#clip-skip-tokens)
|
- [CLIP skip tokens](#clip-skip-tokens)
|
||||||
|
- [Prompt stages](#prompt-stages)
|
||||||
- [Long prompt weighting syntax](#long-prompt-weighting-syntax)
|
- [Long prompt weighting syntax](#long-prompt-weighting-syntax)
|
||||||
- [Pipelines](#pipelines)
|
- [Pipelines](#pipelines)
|
||||||
- [ControlNet pipeline](#controlnet-pipeline)
|
- [ControlNet pipeline](#controlnet-pipeline)
|
||||||
|
@ -424,14 +425,28 @@ than the other tokens and have more parameters, which may change in the future.
|
||||||
<region:top:left:bottom:right:strength:feather:prompt>
|
<region:top:left:bottom:right:strength:feather:prompt>
|
||||||
```
|
```
|
||||||
|
|
||||||
- the prompt can end with +
|
- `top`, `left`, `bottom`, and `right` define the four corners of a rectangle
|
||||||
- top/left/bottom/right are integer pixels
|
- must be integers
|
||||||
- will be rounded down to nearest 8
|
- will be rounded down to the nearest multiple of 8
|
||||||
- strength and feather are floats
|
- `strength` defines the ratio between the two prompts
|
||||||
- strength > 10 replaces other prompt
|
- must be a float or integer
|
||||||
- strength < 0 does weird things
|
- strength should be between 0.0 and 10.0
|
||||||
- feather can be 0.0 to 0.5, reasonably
|
- 2.0 to 5.0 generally works
|
||||||
- prompt can end with + to append the regular prompt
|
- 10.0 completely replaces the base prompt
|
||||||
|
- < 0 does weird things
|
||||||
|
- `feather` defines the blending between the two prompts
|
||||||
|
- must be a float or integer
|
||||||
|
- this is similar to UNet and VAE overlap
|
||||||
|
- feather should be between 0.0 and 0.5
|
||||||
|
- 0.0 will cause hard edges
|
||||||
|
- 0.25 is a good default
|
||||||
|
- the region has its own `prompt`
|
||||||
|
- any characters _except_ `>`
|
||||||
|
- if the region prompt ends with `+`, the base prompt will be appended to it
|
||||||
|
- this can help the region blend with the rest of the image better
|
||||||
|
- `<region:0:0:1024:1024:5.0:0.25:small dog,+> autumn forest, detailed background, 4k, HDR` will use two prompts:
|
||||||
|
- `small dog, autumn forest, detailed background, 4k, HDR` for the region
|
||||||
|
- `autumn forest, detailed background, 4k, HDR` for the rest of the image
|
||||||
|
|
||||||
#### CLIP skip tokens
|
#### CLIP skip tokens
|
||||||
|
|
||||||
|
@ -443,6 +458,10 @@ You can skip the last layers of the CLIP text encoder using the `clip` token:
|
||||||
|
|
||||||
This makes your prompt less specific and some models have been trained to work better with some amount of skipping.
|
This makes your prompt less specific and some models have been trained to work better with some amount of skipping.
|
||||||
|
|
||||||
|
#### Prompt stages
|
||||||
|
|
||||||
|
TODO: explain `first stage || hires prompt` syntax
|
||||||
|
|
||||||
### Long prompt weighting syntax
|
### Long prompt weighting syntax
|
||||||
|
|
||||||
You can emphasize or deemphasize certain parts of the prompt by using the long prompt weighting option. This adds
|
You can emphasize or deemphasize certain parts of the prompt by using the long prompt weighting option. This adds
|
||||||
|
|
Loading…
Reference in New Issue