This extension is an extension of the built-in Composable Diffusion. This allows you to determine the region of the latent space that reflects your subprompts.
The effect of Latent Couple appears only when Enabled is checked.
Regions are created based on these parameters.
Divisions are separated by commas, and the number of divisions is
denoted in the format V:H where V and H is the number of parts a region is
divided along the vertical and horizontal axes, respectively.
Example: 1:1,2:2,5:4
The positions parameter follows a similar format, but the format is V_start-V_end:H_start-H_end
where V_start-V_end denotes the range of positions the prompt affects.
The positions are zero-indexed, meaning 0 is the first position, 1 the second, etc. They go from top to bottom and left to right, so 0:0 is the top left corner in a 2:2 division, and 1:1 is the bottom right corner
The end position is non-inclusive and can be omitted, so '0:0' is equivalent to 0-1:0-1, 1:1 is equivalent to 1-2:1-2, etc.
the positions of different prompts can overlap.
Example: 0:0,1:1,1-4:1-3
Weights are simply the strength for the region in case it overlaps with other regions. It can be any value from -1 to 1
Example: 0.2,0.5,-0.5
The processing of Latent Couple is performed until the specified step is reached.
Press the Visualize button and you will get an image grid that visualizes the regions represented by the parameters.
When you press the Send to button in PNG Info, the parameters of the saved Latent Couple are set to Extra generation params. Press the Apply button to actually apply the parameters.
Extra generation params provided by this extension are saved as PNG Info in the output file.
Latent Couple: "divisions=1:1,1:2,1:2 positions=0:0,0:0,0:1 weights=0.2,0.8,0.8 end at step=20"
parameters
((ultra-detailed)), ((illustration)), 2girls
AND ((ultra-detailed)), ((illustration)), 2girls, black hair
AND ((ultra-detailed)), ((illustration)), 2girls, blonde hair
Negative prompt: (low quality, worst quality:1.4)
Steps: 28, Sampler: Euler a, CFG scale: 5, Seed: 722014241, Size: 512x512, Model hash: 3f64f3baf5, Model: Anything-v3.0-fp16, ENSD: 31337, Latent Couple: "divisions=1:1,1:2,1:2 positions=0:0,0:0,0:1 weights=0.2,0.8,0.8 end at step=20", Eta: 0.67
outputs
- end at step=20 https://imgur.com/OEaRFgh
- end at step=4 https://imgur.com/a1kyvhX
- end at step=0 https://imgur.com/yhGF7g8
This fix is no longer suitable for latest webui commit at 22bcc7be, with gradio dependency upgraded to 3.23.
I'll keep the fix here for people still using older versions of webui.
Activate your venv in webui root directory
For Windows, in cmd
venv\Scripts\activate.bat
For Linux
source venv/bin/activate
Then, install wheel distribution with bugfix applied
pip install --force-reinstall --no-deps extensions/stable-diffusion-webui-two-shot/gradio-3.16.2-py3-none-any.whl
For bugfix related modifications, see https://github.com/ashen-sensored/gradio/tree/3.16.2
The extension's mask color sketching function does not work well with chrome(extreme stuttering) due to gradio's Image component bug.Please keep the browser scaling at 100% while creating blank canvas to avoid the bug. See prerequisite above. The fix is no longer suitable for latest webui version at 22bcc7be, with gradio dependency upgraded to 3.23.
- two shot diffusion.ipynb https://colab.research.google.com/drive/1UdElpQfKFjY5luch9v_LlmSdH7AmeiDe?usp=sharing