SD3 Text Conditioning With Options. One Prompt (Mikey)¶
Documentation¶
- Class name:
SD3TextConditioningWithOptionsOnePrompt - Category:
Mikey/SD3/Conditioning - Output node:
False
The SD3TextConditioningWithOptionsOnePrompt node is designed to enhance text prompts for image generation by applying style-based modifications and conditioning. It leverages specific styles to modify positive and negative prompts, integrates seed-based variations for style selection, and utilizes advanced text encoding techniques to refine the conditioning process for improved image generation outcomes.
Input types¶
Required¶
positive_prompt- Represents the base positive prompt that will be modified according to the selected style and further processed to enhance its effectiveness in guiding the image generation process.
- Comfy dtype:
STRING - Python dtype:
str
negative_prompt- Serves as the base negative prompt that undergoes style-based modifications and processing to refine its role in the image generation process.
- Comfy dtype:
STRING - Python dtype:
str
clip- Specifies the CLIP model to be used for encoding the prompts, playing a crucial role in the conditioning process.
- Comfy dtype:
CLIP - Python dtype:
torch.nn.Module
option_positive_clip_l- Option to apply specific CLIP model settings for the positive prompt's local features.
- Comfy dtype:
COMBO[STRING] - Python dtype:
dict
option_positive_clip_g- Option to apply specific CLIP model settings for the positive prompt's global features.
- Comfy dtype:
COMBO[STRING] - Python dtype:
dict
option_positive_t5xxl- Option to apply specific T5-XXL model settings for the positive prompt.
- Comfy dtype:
COMBO[STRING] - Python dtype:
dict
option_negative_clip_l- Option to apply specific CLIP model settings for the negative prompt's local features.
- Comfy dtype:
COMBO[STRING] - Python dtype:
dict
option_negative_clip_g- Option to apply specific CLIP model settings for the negative prompt's global features.
- Comfy dtype:
COMBO[STRING] - Python dtype:
dict
option_negative_t5xxl- Option to apply specific T5-XXL model settings for the negative prompt.
- Comfy dtype:
COMBO[STRING] - Python dtype:
dict
padding_character- Defines the character used for padding in the text encoding process, affecting the alignment and spacing within the encoded prompts.
- Comfy dtype:
STRING - Python dtype:
str
Output types¶
positive_conditioning- Comfy dtype:
CONDITIONING - The modified positive prompt, enhanced with style-based modifications and advanced encoding techniques for improved image generation.
- Python dtype:
str
- Comfy dtype:
negative_conditioning- Comfy dtype:
CONDITIONING - The enhanced negative prompt, similarly modified with style-based adjustments and encoding to refine the image generation process.
- Python dtype:
str
- Comfy dtype:
Usage tips¶
- Infra type:
CPU - Common nodes: unknown
Source code¶
class SD3TextConditioningWithOptionsOnePrompt:
@classmethod
def INPUT_TYPES(cls):
return {'required': {'positive_prompt': ('STRING', {'default': 'positive prompt', 'multiline': True}),
'negative_prompt': ('STRING', {'default': 'negative prompt', 'multiline': True}),
'clip': ('CLIP',),
'option_positive_clip_l': (['Unmodified','Padded','Empty'], {'default': 'Unmodified'}),
'option_positive_clip_g': (['Unmodified','Padded','Empty'], {'default': 'Unmodified'}),
'option_positive_t5xxl': (['Unmodified','Padded','Empty'], {'default': 'Unmodified'}),
'option_negative_clip_l': (['Unmodified','Padded','Empty'], {'default': 'Unmodified'}),
'option_negative_clip_g': (['Unmodified','Padded','Empty'], {'default': 'Unmodified'}),
'option_negative_t5xxl': (['Unmodified','Padded','Empty'], {'default': 'Unmodified'}),
'padding_character': ('STRING', {'default': ','})}}
RETURN_TYPES = ('CONDITIONING','CONDITIONING',)
RETURN_NAMES = ('positive_conditioning','negative_conditioning')
FUNCTION = 'process'
CATEGORY = 'Mikey/SD3/Conditioning'
def process(self, positive_prompt, negative_prompt, clip, option_positive_clip_l, option_positive_clip_g, option_positive_t5xxl, option_negative_clip_l, option_negative_clip_g, option_negative_t5xxl, padding_character):
positive_tokens = clip.tokenize(positive_prompt)
if len(positive_prompt) == 0 or option_positive_clip_g == 'Empty':
positive_tokens['g'] = []
if len(positive_tokens['g']) < 77 and option_positive_clip_l == 'Padded':
padded_g_string = positive_prompt + padding_character * (77 - len(positive_tokens['g']))
positive_tokens['g'] = clip.tokenize(padded_g_string)['g']
if len(positive_prompt) == 0 or option_positive_clip_l == 'Empty':
positive_tokens['l'] = []
elif len(positive_tokens['l']) < 77 and option_positive_clip_l == 'Padded':
padded_l_string = positive_prompt + padding_character * (77 - len(positive_tokens['l']))
positive_tokens['l'] = clip.tokenize(padded_l_string)['l']
else:
positive_tokens['l'] = clip.tokenize(positive_prompt)['l']
if len(positive_prompt) == 0 or option_positive_t5xxl == 'Empty':
positive_tokens['t5xxl'] = []
elif len(positive_tokens['t5xxl']) < 512 and option_positive_t5xxl == 'Padded':
padded_t5xxl_string = positive_prompt + padding_character * (512 - len(positive_tokens['t5xxl']))
positive_tokens['t5xxl'] = clip.tokenize(padded_t5xxl_string)['t5xxl']
else:
positive_tokens['t5xxl'] = clip.tokenize(positive_prompt)['t5xxl']
if len(positive_tokens['l']) != len(positive_tokens['g']):
empty = clip.tokenize('')
while len(positive_tokens['l']) < len(positive_tokens['g']):
positive_tokens['l'] += empty['l']
while len(positive_tokens['l']) > len(positive_tokens['g']):
positive_tokens['g'] += empty['g']
positive_conditioning, positive_pooled = clip.encode_from_tokens(positive_tokens, return_pooled=True)
# now negative
negative_tokens = clip.tokenize(negative_prompt)
if len(negative_prompt) == 0 or option_negative_clip_g == 'Empty':
negative_tokens['g'] = []
if len(negative_tokens['g']) < 77 and option_negative_clip_l == 'Padded':
padded_g_string = negative_prompt + padding_character * (77 - len(negative_tokens['g']))
negative_tokens['g'] = clip.tokenize(padded_g_string)['g']
if len(negative_prompt) == 0 or option_negative_clip_l == 'Empty':
negative_tokens['l'] = []
elif len(negative_tokens['l']) < 77 and option_negative_clip_l == 'Padded':
padded_l_string = negative_prompt + padding_character * (75 - len(negative_tokens['l']))
negative_tokens['l'] = clip.tokenize(padded_l_string)['l']
else:
negative_tokens['l'] = clip.tokenize(negative_prompt)['l']
if len(negative_prompt) == 0 or option_negative_t5xxl == 'Empty':
negative_tokens['t5xxl'] = []
elif len(negative_tokens['t5xxl']) < 512 and option_negative_t5xxl == 'Padded':
padded_t5xxl_string = negative_prompt + padding_character * (512 - len(negative_tokens['t5xxl']))
negative_tokens['t5xxl'] = clip.tokenize(padded_t5xxl_string)['t5xxl']
else:
negative_tokens['t5xxl'] = clip.tokenize(negative_prompt)['t5xxl']
if len(negative_tokens['l']) != len(negative_tokens['g']):
empty = clip.tokenize('')
while len(negative_tokens['l']) < len(negative_tokens['g']):
negative_tokens['l'] += empty['l']
while len(negative_tokens['l']) > len(negative_tokens['g']):
negative_tokens['g'] += empty['g']
negative_conditioning, negative_pooled = clip.encode_from_tokens(negative_tokens, return_pooled=True)
return (
[[positive_conditioning, {'pooled_output': positive_pooled}]],
[[negative_conditioning, {'pooled_output': negative_pooled}]],
)