v13c CLSP SDEdit

Source

Prompt

CLSP CFG 1.00 ? Amount to which prompt conditioning is applied to the denoising process.
Workable values are usually around 0.5 to 3. CLSP Amplification 1.00 ? Extremity of the prompt vector itself.
Values above 1 move it outside of the range it was trained on, but sometimes work a bit better. SDEdit Edit Depth 25%anchor 75% ? The bigger this value, the less of the input audio we retain. Bridge Mel Mean 0.50 ? Experiment with this value to see how the sound in this particular model of Ruchey is affected by the amount of noise injected into the mel-spectrogram before denoising.
There is an optimal value for every piece of audio, but not a single value for all.
This is an artifact of this version and will be removed. Bridge Mel Std 5.00 ? Same as Mel Mean, this affects the amount of noise injected into the mel-spectrogram, and it affects the output quality.
There is an optimal value for every piece of audio, but not a single value for all.

Externally preserve source identity ? If this is on, we take the identity vectors from the input and apply them to the output.
In case the prompt affects identity, like with whispering, it may stop doing that. Externally preserve source pitch ? If this is on, the pitch generated by the Transformer model will not be used.
If the prompt had to change pitch, it will not affect the output anymore. Externally preserve source energy ? If this is on, the energy generated by the Transformer model will not be used.
If the prompt had to change cadence and rhythm, it will not affect the output anymore, or will affect it unevenly. Normalize generated pitch mean/std ? Normalizes generated pitch because it is often generated in the wrong octave.
This is an artifact of this Transformer v13c model.

Result

No generation yet