K00B404's picture
Create app.py
8ac3922 verified
import gradio as gr
import torch
import torch.nn.functional as F
from diffusers import StableDiffusionImg2ImgPipeline, DDIMScheduler
from PIL import Image
import numpy as np
from typing import List, Optional, Dict, Any
from collections import deque
import cv2
import os
import tempfile
import imageio
from datetime import datetime
class SimpleTemporalBuffer:
"""Simplified temporal buffer for SD1.5 img2img"""
def __init__(self, buffer_size: int = 6):
self.buffer_size = buffer_size
self.frames = deque(maxlen=buffer_size)
self.frame_embeddings = deque(maxlen=buffer_size)
self.motion_vectors = deque(maxlen=buffer_size-1)
def add_frame(self, frame: Image.Image, embedding: Optional[torch.Tensor] = None):
"""Add frame to buffer"""
try:
# Calculate optical flow if we have previous frames
if len(self.frames) > 0:
prev_frame = np.array(self.frames[-1])
curr_frame = np.array(frame)
# Convert to grayscale for optical flow
prev_gray = cv2.cvtColor(prev_frame, cv2.COLOR_RGB2GRAY)
curr_gray = cv2.cvtColor(curr_frame, cv2.COLOR_RGB2GRAY)
# Calculate optical flow
flow = cv2.calcOpticalFlowPyrLK(
prev_gray, curr_gray,
np.array([[frame.width//2, frame.height//2]], dtype=np.float32),
None
)[0]
if flow is not None:
motion_magnitude = np.linalg.norm(flow[0] - [frame.width//2, frame.height//2])
self.motion_vectors.append(motion_magnitude)
except Exception as e:
print(f"Motion calculation error: {e}")
self.frames.append(frame)
if embedding is not None:
self.frame_embeddings.append(embedding)
def get_reference_frame(self) -> Optional[Image.Image]:
"""Get most recent frame as reference"""
return self.frames[-1] if self.frames else None
def get_motion_context(self) -> Dict[str, Any]:
"""Get motion context for next frame generation"""
if len(self.motion_vectors) == 0:
return {"has_motion": False, "predicted_motion": 0.0}
# Simple motion prediction based on recent vectors
recent_motion = list(self.motion_vectors)[-3:] # Last 3 motions
avg_motion = np.mean(recent_motion)
motion_trend = recent_motion[-1] - recent_motion[0] if len(recent_motion) > 1 else 0
predicted_motion = avg_motion + motion_trend * 0.5
return {
"has_motion": True,
"current_motion": avg_motion,
"predicted_motion": predicted_motion,
"motion_trend": motion_trend,
"motion_history": recent_motion
}
class SD15FlexibleI2VGenerator:
"""Flexible I2V generator using SD1.5 img2img pipeline"""
def __init__(
self,
model_id: str = "runwayml/stable-diffusion-v1-5",
device: str = "cuda" if torch.cuda.is_available() else "cpu"
):
self.device = device
self.pipe = None
self.temporal_buffer = SimpleTemporalBuffer()
self.is_loaded = False
def load_model(self):
"""Load the SD1.5 pipeline"""
if self.is_loaded:
return "Model already loaded"
try:
print(f"πŸš€ Loading SD1.5 pipeline on {self.device}...")
# Load pipeline with DDIM scheduler for better img2img
self.pipe = StableDiffusionImg2ImgPipeline.from_pretrained(
"runwayml/stable-diffusion-v1-5",
torch_dtype=torch.float16 if self.device == "cuda" else torch.float32,
safety_checker=None,
requires_safety_checker=False
)
# Use DDIM for more consistent results
self.pipe.scheduler = DDIMScheduler.from_config(self.pipe.scheduler.config)
self.pipe = self.pipe.to(self.device)
# Enable memory efficient attention
if self.device == "cuda":
self.pipe.enable_attention_slicing()
try:
self.pipe.enable_xformers_memory_efficient_attention()
except:
print("⚠️ xformers not available, using standard attention")
self.is_loaded = True
return "βœ… Model loaded successfully!"
except Exception as e:
return f"❌ Error loading model: {str(e)}"
def calculate_adaptive_strength(self, motion_context: Dict[str, Any], base_strength: float = 0.75) -> float:
"""Calculate adaptive denoising strength based on motion"""
if not motion_context.get("has_motion", False):
return base_strength
motion = motion_context["current_motion"]
# More motion = less strength (preserve more of previous frame)
# Less motion = more strength (allow more change)
motion_factor = np.clip(motion / 50.0, 0.0, 1.0) # Normalize motion
adaptive_strength = base_strength * (1.0 - motion_factor * 0.3)
return np.clip(adaptive_strength, 0.3, 0.9)
def enhance_prompt_with_motion(self, base_prompt: str, motion_context: Dict[str, Any]) -> str:
"""Enhance prompt based on motion context"""
if not motion_context.get("has_motion", False):
return base_prompt
motion = motion_context["current_motion"]
trend = motion_context.get("motion_trend", 0)
# Add motion descriptors based on analysis
if motion > 30:
if trend > 5:
motion_desc = ", fast movement, dynamic motion, motion blur"
else:
motion_desc = ", steady movement, continuous motion"
elif motion > 10:
motion_desc = ", gentle movement, smooth transition"
else:
motion_desc = ", subtle movement, slight change"
return base_prompt + motion_desc
def blend_frames(self, current_frame: Image.Image, reference_frame: Image.Image, blend_ratio: float = 0.15) -> Image.Image:
"""Blend current frame with reference for temporal consistency"""
current_array = np.array(current_frame, dtype=np.float32)
reference_array = np.array(reference_frame, dtype=np.float32)
# Blend frames
blended_array = current_array * (1 - blend_ratio) + reference_array * blend_ratio
blended_array = np.clip(blended_array, 0, 255).astype(np.uint8)
return Image.fromarray(blended_array)
@torch.no_grad()
def generate_frame_batch(
self,
init_image: Image.Image,
prompt: str,
num_frames: int = 1,
strength: float = 0.75,
guidance_scale: float = 7.5,
num_inference_steps: int = 20,
generator: Optional[torch.Generator] = None,
progress_callback=None
) -> List[Image.Image]:
"""Generate a batch of frames using img2img"""
if not self.is_loaded:
raise ValueError("Model not loaded. Please load the model first.")
frames = []
current_image = init_image
for i in range(num_frames):
if progress_callback:
progress_callback(f"Generating frame {i+1}/{num_frames}")
# Get motion context
motion_context = self.temporal_buffer.get_motion_context()
# Adaptive parameters based on motion
adaptive_strength = self.calculate_adaptive_strength(motion_context, strength)
enhanced_prompt = self.enhance_prompt_with_motion(prompt, motion_context)
# Generate frame
result = self.pipe(
prompt=enhanced_prompt,
image=current_image,
strength=adaptive_strength,
guidance_scale=guidance_scale,
num_inference_steps=num_inference_steps,
generator=generator
)
generated_frame = result.images[0]
# Apply temporal consistency blending
if len(self.temporal_buffer.frames) > 0:
reference_frame = self.temporal_buffer.get_reference_frame()
blend_ratio = 0.1 if motion_context.get("current_motion", 0) > 20 else 0.2
generated_frame = self.blend_frames(generated_frame, reference_frame, blend_ratio)
# Update buffer
self.temporal_buffer.add_frame(generated_frame)
frames.append(generated_frame)
# Use generated frame as input for next iteration
current_image = generated_frame
return frames
def generate_i2v_sequence(
self,
init_image: Image.Image,
prompt: str,
total_frames: int = 16,
frames_per_batch: int = 2,
strength: float = 0.75,
guidance_scale: float = 7.5,
num_inference_steps: int = 20,
seed: Optional[int] = None,
progress_callback=None
) -> List[Image.Image]:
"""Generate I2V sequence with flexible batch sizes"""
if not self.is_loaded:
raise ValueError("Model not loaded. Please load the model first.")
# Setup generator
generator = torch.Generator(device=self.device)
if seed is not None:
generator.manual_seed(seed)
# Reset temporal buffer and add initial frame
self.temporal_buffer = SimpleTemporalBuffer()
self.temporal_buffer.add_frame(init_image)
all_frames = [init_image] # Start with initial frame
frames_generated = 1
current_reference = init_image
# Generate in batches
while frames_generated < total_frames:
remaining_frames = total_frames - frames_generated
current_batch_size = min(frames_per_batch, remaining_frames)
if progress_callback:
progress_callback(f"Batch: Generating frames {frames_generated+1}-{frames_generated+current_batch_size}")
# Generate batch
batch_frames = self.generate_frame_batch(
init_image=current_reference,
prompt=prompt,
num_frames=current_batch_size,
strength=strength,
guidance_scale=guidance_scale,
num_inference_steps=num_inference_steps,
generator=generator,
progress_callback=progress_callback
)
# Add to results
all_frames.extend(batch_frames)
frames_generated += current_batch_size
# Update reference for next batch
current_reference = batch_frames[-1]
return all_frames
# Global generator instance
generator = SD15FlexibleI2VGenerator()
def load_model_interface():
"""Interface function to load the model"""
status = generator.load_model()
return status
def create_frames_to_gif(frames: List[Image.Image], duration: int = 200) -> str:
"""Convert frames to GIF and return file path"""
temp_dir = tempfile.mkdtemp()
timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
gif_path = os.path.join(temp_dir, f"i2v_sequence_{timestamp}.gif")
frames[0].save(
gif_path,
save_all=True,
append_images=frames[1:],
duration=duration,
loop=0
)
return gif_path
def create_frames_to_video(frames: List[Image.Image], fps: int = 8) -> str:
"""Convert frames to MP4 video and return file path"""
temp_dir = tempfile.mkdtemp()
timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
video_path = os.path.join(temp_dir, f"i2v_sequence_{timestamp}.mp4")
try:
with imageio.get_writer(video_path, fps=fps) as writer:
for frame in frames:
writer.append_data(np.array(frame))
return video_path
except ImportError:
# Fallback to GIF if imageio not available
return create_frames_to_gif(frames, duration=int(1000/fps))
def generate_i2v_interface(
init_image,
prompt,
total_frames,
frames_per_batch,
strength,
guidance_scale,
num_inference_steps,
seed,
output_format,
progress=gr.Progress()
):
"""Main interface function for I2V generation"""
if init_image is None:
return None, None, "❌ Please upload an initial image"
if not prompt.strip():
return None, None, "❌ Please enter a prompt"
try:
# Progress callback
def update_progress(message):
progress(0.5, desc=message)
progress(0.1, desc="Starting generation...")
# Resize image to 512x512 if needed
if init_image.size != (512, 512):
init_image = init_image.resize((512, 512), Image.Resampling.LANCZOS)
# Generate frames
frames = generator.generate_i2v_sequence(
init_image=init_image,
prompt=prompt,
total_frames=total_frames,
frames_per_batch=frames_per_batch,
strength=strength,
guidance_scale=guidance_scale,
num_inference_steps=num_inference_steps,
seed=seed if seed > 0 else None,
progress_callback=update_progress
)
progress(0.8, desc="Creating output file...")
# Create output file
if output_format == "GIF":
output_path = create_frames_to_gif(frames, duration=200)
else: # MP4
output_path = create_frames_to_video(frames, fps=8)
progress(1.0, desc="Complete!")
# Return last frame as preview and the output file
return frames[-1], output_path, f"βœ… Generated {len(frames)} frames successfully!"
except Exception as e:
return None, None, f"❌ Error: {str(e)}"
def generate_variable_pattern_interface(
init_image,
prompt,
total_frames,
batch_pattern_str,
strength,
guidance_scale,
num_inference_steps,
seed,
output_format,
progress=gr.Progress()
):
"""Interface for variable batch pattern generation"""
if init_image is None:
return None, None, "❌ Please upload an initial image"
if not prompt.strip():
return None, None, "❌ Please enter a prompt"
try:
# Parse batch pattern
batch_pattern = [int(x.strip()) for x in batch_pattern_str.split(",")]
if not batch_pattern or any(x <= 0 for x in batch_pattern):
raise ValueError("Invalid batch pattern")
progress(0.1, desc="Starting variable pattern generation...")
# Resize image
if init_image.size != (512, 512):
init_image = init_image.resize((512, 512), Image.Resampling.LANCZOS)
# Generate with variable pattern
frames = [init_image]
frames_generated = 1
current_reference = init_image
pattern_idx = 0
generator.temporal_buffer = SimpleTemporalBuffer()
generator.temporal_buffer.add_frame(init_image)
gen = torch.Generator(device=generator.device)
if seed > 0:
gen.manual_seed(seed)
while frames_generated < total_frames:
current_batch_size = batch_pattern[pattern_idx % len(batch_pattern)]
remaining_frames = total_frames - frames_generated
actual_batch_size = min(current_batch_size, remaining_frames)
progress(frames_generated / total_frames,
desc=f"Pattern step {pattern_idx+1}: {actual_batch_size} frames")
batch_frames = generator.generate_frame_batch(
init_image=current_reference,
prompt=prompt,
num_frames=actual_batch_size,
strength=strength,
guidance_scale=guidance_scale,
num_inference_steps=num_inference_steps,
generator=gen
)
frames.extend(batch_frames)
frames_generated += actual_batch_size
current_reference = batch_frames[-1]
pattern_idx += 1
progress(0.9, desc="Creating output file...")
# Create output
final_frames = frames[:total_frames+1] # Include initial frame
if output_format == "GIF":
output_path = create_frames_to_gif(final_frames, duration=200)
else:
output_path = create_frames_to_video(final_frames, fps=8)
progress(1.0, desc="Complete!")
return final_frames[-1], output_path, f"βœ… Generated {len(final_frames)} frames with pattern {batch_pattern}!"
except Exception as e:
return None, None, f"❌ Error: {str(e)}"
# Create Gradio interface
def create_gradio_app():
"""Create the main Gradio application"""
with gr.Blocks(title="SD1.5 Flexible I2V Generator", theme=gr.themes.Soft()) as app:
gr.Markdown("""
# 🎬 SD1.5 Flexible I2V Generator
Generate image-to-video sequences with **flexible batch processing** and **temporal consistency**!
## Key Features:
- 🎯 **Flexible Batch Sizes**: Generate 1, 2, 3+ frames at a time
- πŸ”„ **Motion-Aware Processing**: Adapts based on detected motion
- 🎨 **Temporal Consistency**: Smooth transitions between frames
- πŸ“ˆ **Variable Patterns**: Dynamic batch sizing patterns
""")
# Model loading section
with gr.Row():
load_btn = gr.Button("πŸš€ Load SD1.5 Model", variant="primary", size="lg")
model_status = gr.Textbox(
label="Model Status",
value="Model not loaded. Click 'Load SD1.5 Model' to start.",
interactive=False
)
load_btn.click(load_model_interface, outputs=model_status)
# Main interface tabs
with gr.Tabs():
# Fixed batch size tab
with gr.Tab("🎯 Fixed Batch Generation"):
with gr.Row():
with gr.Column(scale=1):
init_image_1 = gr.Image(
label="Initial Image",
type="pil",
height=300
)
prompt_1 = gr.Textbox(
label="Prompt",
placeholder="e.g., a cat walking through a peaceful garden, cinematic lighting",
lines=3
)
with gr.Row():
total_frames_1 = gr.Slider(
label="Total Frames",
minimum=4,
maximum=32,
value=12,
step=1
)
frames_per_batch_1 = gr.Slider(
label="Frames per Batch (Key Parameter!)",
minimum=1,
maximum=4,
value=2,
step=1
)
with gr.Accordion("Advanced Settings", open=False):
strength_1 = gr.Slider(
label="Strength",
minimum=0.3,
maximum=0.9,
value=0.75,
step=0.05
)
guidance_scale_1 = gr.Slider(
label="Guidance Scale",
minimum=3.0,
maximum=15.0,
value=7.5,
step=0.5
)
num_inference_steps_1 = gr.Slider(
label="Inference Steps",
minimum=10,
maximum=50,
value=20,
step=5
)
seed_1 = gr.Number(
label="Seed (-1 for random)",
value=-1
)
output_format_1 = gr.Radio(
label="Output Format",
choices=["GIF", "MP4"],
value="GIF"
)
generate_btn_1 = gr.Button("🎬 Generate I2V Sequence", variant="primary", size="lg")
with gr.Column(scale=1):
preview_1 = gr.Image(label="Last Frame Preview", height=300)
output_file_1 = gr.File(label="Download Generated Video/GIF")
status_1 = gr.Textbox(label="Status", interactive=False)
generate_btn_1.click(
generate_i2v_interface,
inputs=[
init_image_1, prompt_1, total_frames_1, frames_per_batch_1,
strength_1, guidance_scale_1, num_inference_steps_1, seed_1, output_format_1
],
outputs=[preview_1, output_file_1, status_1]
)
# Variable pattern tab
with gr.Tab("πŸ“ˆ Variable Pattern Generation"):
with gr.Row():
with gr.Column(scale=1):
init_image_2 = gr.Image(
label="Initial Image",
type="pil",
height=300
)
prompt_2 = gr.Textbox(
label="Prompt",
placeholder="e.g., smooth camera movement through a scene",
lines=3
)
total_frames_2 = gr.Slider(
label="Total Frames",
minimum=6,
maximum=40,
value=16,
step=1
)
batch_pattern_2 = gr.Textbox(
label="Batch Pattern (comma-separated)",
value="1,2,3,2,1",
placeholder="e.g., 1,2,3,2,1 or 2,4,2"
)
gr.Markdown("""
**Pattern Examples:**
- `1,2,3,2,1` - Start slow, ramp up, slow down
- `2,2,2,2` - Consistent 2-frame batches
- `1,3,1,3` - Alternating single and triple
""")
with gr.Accordion("Advanced Settings", open=False):
strength_2 = gr.Slider(label="Strength", minimum=0.3, maximum=0.9, value=0.75, step=0.05)
guidance_scale_2 = gr.Slider(label="Guidance Scale", minimum=3.0, maximum=15.0, value=7.5, step=0.5)
num_inference_steps_2 = gr.Slider(label="Inference Steps", minimum=10, maximum=50, value=20, step=5)
seed_2 = gr.Number(label="Seed (-1 for random)", value=-1)
output_format_2 = gr.Radio(label="Output Format", choices=["GIF", "MP4"], value="GIF")
generate_btn_2 = gr.Button("🎨 Generate with Pattern", variant="primary", size="lg")
with gr.Column(scale=1):
preview_2 = gr.Image(label="Last Frame Preview", height=300)
output_file_2 = gr.File(label="Download Generated Video/GIF")
status_2 = gr.Textbox(label="Status", interactive=False)
generate_btn_2.click(
generate_variable_pattern_interface,
inputs=[
init_image_2, prompt_2, total_frames_2, batch_pattern_2,
strength_2, guidance_scale_2, num_inference_steps_2, seed_2, output_format_2
],
outputs=[preview_2, output_file_2, status_2]
)
# Examples section
with gr.Accordion("πŸ“ Example Prompts & Tips", open=False):
gr.Markdown("""
## 🎯 Good Prompts for I2V:
- `a peaceful lake with gentle ripples, soft sunlight, cinematic`
- `a cat slowly walking through a garden, smooth movement`
- `camera slowly panning across a mountain landscape`
- `a flower blooming in timelapse, natural lighting`
- `gentle waves on a beach, golden hour lighting`
## πŸ›  Parameter Tips:
- **Frames per Batch**:
- `1` = Maximum consistency, slower generation
- `2-3` = Balanced quality and speed
- `4+` = Faster but less consistent
- **Strength**:
- `0.6-0.7` = Subtle changes
- `0.7-0.8` = Moderate animation
- `0.8-0.9` = More dramatic changes
- **Batch Patterns**:
- Use `1,2,3,2,1` for organic acceleration/deceleration
- Use consistent values like `2,2,2` for steady pacing
""")
gr.Markdown("""
---
## πŸš€ **Innovation Highlights:**
This app demonstrates **flexible batch processing** for I2V generation:
- Generate multiple frames simultaneously with `frames_per_batch`
- Motion-aware strength adaptation based on optical flow
- Temporal consistency through intelligent frame blending
- Variable stepping patterns for dynamic control
**Built with SD1.5 img2img pipeline + custom temporal processing!**
""")
return app
if __name__ == "__main__":
app = create_gradio_app()
app.launch(
server_name="0.0.0.0",
server_port=7860,
share=False,
debug=True
)