Wanxiang XL_Ultra Realistic Photography

Note: [from:to:when] commands , which are used in the gallery examples , no longer work on TensorArt for TAMS2 .

//---//

Please credit the original author: https://www.shakker.ai/modelinfo/b27d7c6bd9184acfbfefd2a89909c6d3?form=pic_detail

Recommended config for [from:to:steps] (Preliminary)

Steps: 25

CFG : 6

Sampler: DPM++ 2M Karras (any will do as long as its non-ancestral)

CLIP Skip : 2

Recommended config for straight prompting (By model author):

Steps: 25-30

CFG : 7

Sampler: DPM++ 2S a

CLIP Skip : 2

Negatives:

Pick 3 tokens at random , then activate past 50% of generation steps , for example

NEG: [:sasia harbaugh judo : 0.7]

Exceptions:

When using very exotic tokens like emojis , switch from CLIP Skip 2 to CLIP skip 1

About the model:

• It's good at creating shadows.

• Not built for NSFW content

• Slight Asian bias

• Struggles with skin texture compared to SD 1.5 variants.