comi-logic/Wan2GP
7
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

LTXV and Flux updates

Browse Source
This commit is contained in:
deepbeepmeep
2025-07-21 09:37:52 +02:00
parent a356c6af4b
commit a1c228054c
36 changed files with 1285 additions and 2288 deletions
Download Patch File Download Diff File Expand all files Collapse all files

40
ltx_video/utils/prompt_enhance_utils.py
View File

@@ -23,6 +23,23 @@ Note any changes or sudden events
Do not exceed the 150 word limit!
Output the enhanced prompt only.
"""
T2I_VISUAL_PROMPT = """You are an expert visual artist and photographer with award-winning compositions. When writing prompts based on the user input, focus on detailed, precise descriptions of visual elements and composition.
Include specific poses, appearances, framing, and environmental details - all in a single flowing paragraph.
Start directly with the main subject, and keep descriptions literal and precise.
Think like a photographer describing the perfect shot.
Do not change the user input intent, just enhance it.
Keep within 150 words.
For best results, build your prompts using this structure:
Start with main subject and pose in a single sentence
Add specific details about expressions and positioning
Describe character/object appearances precisely
Include background and environment details
Specify framing, composition and perspective
Describe lighting, colors, and mood
Note any atmospheric or stylistic elements
Do not exceed the 150 word limit!
Output the enhanced prompt only.
"""
I2V_CINEMATIC_PROMPT = """You are an expert cinematic director with many award winning movies, When writing prompts based on the user input, focus on detailed, chronological descriptions of actions and scenes.
Include specific movements, appearances, camera angles, and environmental details - all in a single flowing paragraph.
@@ -43,6 +60,24 @@ Do not exceed the 150 word limit!
Output the enhanced prompt only.
"""
I2I_VISUAL_PROMPT = """You are an expert visual artist and photographer with award-winning compositions. When writing prompts based on the user input, focus on detailed, precise descriptions of visual elements and composition.
Include specific poses, appearances, framing, and environmental details - all in a single flowing paragraph.
Start directly with the main subject, and keep descriptions literal and precise.
Think like a photographer describing the perfect shot.
Do not change the user input intent, just enhance it.
Keep within 150 words.
For best results, build your prompts using this structure:
Start with main subject and pose in a single sentence
Add specific details about expressions and positioning
Describe character/object appearances precisely
Include background and environment details
Specify framing, composition and perspective
Describe lighting, colors, and mood
Note any atmospheric or stylistic elements
Do not exceed the 150 word limit!
Output the enhanced prompt only.
"""
def tensor_to_pil(tensor):
# Ensure tensor is in range [-1, 1]
@@ -68,6 +103,7 @@ def generate_cinematic_prompt(
prompt_enhancer_tokenizer,
prompt: Union[str, List[str]],
images: Optional[List] = None,
video_prompt= True,
max_new_tokens: int = 256,
) -> List[str]:
prompts = [prompt] if isinstance(prompt, str) else prompt
@@ -78,7 +114,7 @@ def generate_cinematic_prompt(
prompt_enhancer_tokenizer,
prompts,
max_new_tokens,
T2V_CINEMATIC_PROMPT,
T2V_CINEMATIC_PROMPT if video_prompt else T2I_VISUAL_PROMPT,
)
else:
@@ -90,7 +126,7 @@ def generate_cinematic_prompt(
prompts,
images,
max_new_tokens,
I2V_CINEMATIC_PROMPT,
I2V_CINEMATIC_PROMPT if video_prompt else I2I_VISUAL_PROMPT,
)
return prompts