"Almost had it..."
- These are my new "default" settings for this model (and ONLY for this model). - The LoRA is set to [0.6], and it works supremely well, but isn't actually necessary to generate crisp clean images at lower steps. I also wouldn't recommend setting it higher since doing so oversaturates the image with details that don't match lighting conditions. - Take note of the CFG and steps. The defaults had me on [28] steps of [Euler a] at [11] CFG. It was awful. The author probably forgot to set the defaults. I recommend staying between [13] - [24] steps and using around a [3] CFG. - This prompt is my way of formatting it so it's easy to understand. You literally don't need any of it - you could just put "sexy elf" or whatever and you would not be disappointed - but the last parts describing the lighting were designed for more realistic lighting under darker conditions and are still useful for basically any prompt.
Mostra parametri
- Dimensione
- 768 x 1344
- Passaggi di campionamento
- 15
- Metodo di campionamento
- Euler a
- Scala CFG
- 2.4
- Seme
- 3141592653589793
- Negativo
- lowres, worst quality, bad quality, jpeg artifacts, signature, watermark,
Correzione del viso
HiRes
Commenti