rickyars
c421531a32
i'm trying to improve the tiled diffusion - these changes in node and node advanced are to ensure that final output is always tile width * grid width by tile height * grid height. i'm also starting work on a new tiled diffusion upscaler.
280 lines
12 KiB
Python
280 lines
12 KiB
Python
import torch
|
|
import comfy.sample
|
|
import comfy.controlnet
|
|
from comfy.utils import ProgressBar
|
|
|
|
from .utils import gaussian_blend_tiles, parse_tile_prompts
|
|
from .utils import apply_controlnet_to_conditioning
|
|
|
|
class TiledImageGenerator:
|
|
"""
|
|
ComfyUI node that generates a tiled image composition with overlapping regions
|
|
that serve as seeds for neighboring tiles, using ControlNet Union SDXL for outpainting.
|
|
"""
|
|
|
|
@classmethod
|
|
def INPUT_TYPES(cls):
|
|
import comfy.samplers
|
|
|
|
return {
|
|
"required": {
|
|
"json_tile_prompts": ("STRING", {"multiline": True}),
|
|
"global_positive": ("STRING", {"multiline": True, "default": ""}),
|
|
"global_negative": ("STRING", {"multiline": True, "default": ""}),
|
|
"grid_width": ("INT", {"default": 4, "min": 1, "max": 16}),
|
|
"grid_height": ("INT", {"default": 6, "min": 1, "max": 16}),
|
|
"tile_width": ("INT", {"default": 1024, "min": 256, "max": 2048}),
|
|
"tile_height": ("INT", {"default": 1024, "min": 256, "max": 2048}),
|
|
"overlap_percent": ("FLOAT", {"default": 0.15, "min": 0.05, "max": 0.5, "step": 0.01}),
|
|
"blend_sigma": ("FLOAT", {"default": 0.4, "min": 0.0, "max": 1.0, "step": 0.01}),
|
|
"model": ("MODEL",),
|
|
"clip": ("CLIP",),
|
|
"vae": ("VAE",),
|
|
"controlnet": ("CONTROL_NET",),
|
|
"controlnet_strength": ("FLOAT", {"default": 0.7, "min": 0.0, "max": 1.0, "step": 0.01}),
|
|
"sampler_name": (comfy.samplers.KSampler.SAMPLERS,),
|
|
"scheduler": (comfy.samplers.KSampler.SCHEDULERS,),
|
|
"steps": ("INT", {"default": 20, "min": 1, "max": 100}),
|
|
"cfg": ("FLOAT", {"default": 7.0, "min": 1.0, "max": 20.0, "step": 0.1}),
|
|
"seed": ("INT", {"default": 0, "min": 0, "max": 0xffffffffffffffff, "control_after_generate": True}),
|
|
}
|
|
}
|
|
|
|
RETURN_TYPES = ("IMAGE", "IMAGE")
|
|
RETURN_NAMES = ("composite_image", "individual_tiles")
|
|
FUNCTION = "generate_tiled_image"
|
|
CATEGORY = "image/generation"
|
|
|
|
def generate_tiled_image(self, json_tile_prompts, global_positive, global_negative,
|
|
grid_width, grid_height, tile_width, tile_height,
|
|
overlap_percent, blend_sigma, controlnet, controlnet_strength,
|
|
seed, model, clip, vae, sampler_name, scheduler, steps, cfg):
|
|
"""Generate a tiled image with uniform grown latents for all tiles."""
|
|
|
|
# Parse the JSON tile prompts
|
|
tile_prompts = parse_tile_prompts(json_tile_prompts, grid_width, grid_height)
|
|
|
|
# Calculate overlap in pixels
|
|
overlap_x = int(tile_width * overlap_percent)
|
|
overlap_y = int(tile_height * overlap_percent)
|
|
|
|
# SIMPLE FINAL SIZE CALCULATION - exact grid multiplication
|
|
final_width = tile_width * grid_width
|
|
final_height = tile_height * grid_height
|
|
|
|
print(f"Final output size: {final_width} x {final_height}")
|
|
print(f"Tile size: {tile_width} x {tile_height}")
|
|
print(f"Grid: {grid_width} x {grid_height}")
|
|
print(f"Overlap: {overlap_x} x {overlap_y} pixels ({overlap_percent * 100}%)")
|
|
|
|
# UNIFORM GENERATION CANVAS SIZE for all tiles
|
|
gen_width = ((tile_width + overlap_x + 7) // 8) * 8
|
|
gen_height = ((tile_height + overlap_y + 7) // 8) * 8
|
|
print(f"Uniform generation canvas: {gen_width} x {gen_height}")
|
|
|
|
# Create a blank canvas for the final composite
|
|
final_tensor = torch.ones((1, final_height, final_width, 3), dtype=torch.float32) * 0.5
|
|
|
|
# Storage for individual tiles and full grown tiles
|
|
individual_tensors = [] # For blending (cropped)
|
|
full_grown_tensors = [] # For debugging (full canvas)
|
|
positions = []
|
|
|
|
# Get device
|
|
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
|
|
print(f"Using device: {device}")
|
|
|
|
# Calculate total number of tiles
|
|
total_tiles = grid_width * grid_height
|
|
pbar = ProgressBar(total_tiles)
|
|
|
|
# Generate tiles one by one
|
|
for y in range(grid_height):
|
|
for x in range(grid_width):
|
|
idx = y * grid_width + x
|
|
current_prompt = tile_prompts[idx]
|
|
current_seed = seed + idx
|
|
|
|
# Calculate position in the final image
|
|
final_pos_x = x * tile_width
|
|
final_pos_y = y * tile_height
|
|
|
|
# Create the combined prompt
|
|
combined_prompt = f"{current_prompt} {global_positive}" if global_positive else current_prompt
|
|
print(f"Generating tile ({x + 1},{y + 1}) with prompt: {current_prompt}")
|
|
|
|
# Encode the combined prompt
|
|
pos_tokens = clip.tokenize(combined_prompt)
|
|
pos_cond = clip.encode_from_tokens_scheduled(pos_tokens)
|
|
|
|
# Encode the negative prompt
|
|
neg_tokens = clip.tokenize(global_negative)
|
|
neg_cond = clip.encode_from_tokens_scheduled(neg_tokens)
|
|
|
|
# Create UNIFORM working canvas for ALL tiles
|
|
working_tensor = torch.zeros((1, gen_height, gen_width, 3), dtype=torch.float32)
|
|
outpaint_mask = torch.ones((1, gen_height, gen_width, 1), dtype=torch.float32)
|
|
|
|
# Check if we have neighbors to copy from
|
|
has_left_neighbor = x > 0
|
|
has_top_neighbor = y > 0
|
|
|
|
# COPY OVERLAP REGIONS from final_tensor (only if neighbors exist)
|
|
if has_left_neighbor:
|
|
# Copy left overlap region from previous tile
|
|
source_x = final_pos_x - overlap_x
|
|
if source_x >= 0 and source_x + overlap_x <= final_width:
|
|
source_start_y = final_pos_y
|
|
source_end_y = min(final_pos_y + tile_height, final_height)
|
|
source_height = source_end_y - source_start_y
|
|
|
|
# Copy to working tensor left edge (after top overlap region)
|
|
target_start_y = overlap_y
|
|
copy_height = min(source_height, gen_height - target_start_y)
|
|
|
|
if copy_height > 0:
|
|
working_tensor[0,
|
|
target_start_y:target_start_y + copy_height,
|
|
:overlap_x,
|
|
:] = final_tensor[0,
|
|
source_start_y:source_start_y + copy_height,
|
|
source_x:source_x + overlap_x,
|
|
:]
|
|
# Mark as "keep" in mask
|
|
outpaint_mask[0,
|
|
target_start_y:target_start_y + copy_height,
|
|
:overlap_x,
|
|
:] = 0
|
|
|
|
if has_top_neighbor:
|
|
# Copy top overlap region from previous tile
|
|
source_y = final_pos_y - overlap_y
|
|
if source_y >= 0 and source_y + overlap_y <= final_height:
|
|
source_start_x = final_pos_x
|
|
source_end_x = min(final_pos_x + tile_width, final_width)
|
|
source_width = source_end_x - source_start_x
|
|
|
|
# Copy to working tensor top edge (after left overlap region)
|
|
target_start_x = overlap_x
|
|
copy_width = min(source_width, gen_width - target_start_x)
|
|
|
|
if copy_width > 0:
|
|
working_tensor[0,
|
|
:overlap_y,
|
|
target_start_x:target_start_x + copy_width,
|
|
:] = final_tensor[0,
|
|
source_y:source_y + overlap_y,
|
|
source_start_x:source_start_x + copy_width,
|
|
:]
|
|
# Mark as "keep" in mask
|
|
outpaint_mask[0,
|
|
:overlap_y,
|
|
target_start_x:target_start_x + copy_width,
|
|
:] = 0
|
|
|
|
# Handle corner overlap if both neighbors exist
|
|
if has_left_neighbor and has_top_neighbor:
|
|
corner_source_x = final_pos_x - overlap_x
|
|
corner_source_y = final_pos_y - overlap_y
|
|
if (corner_source_x >= 0 and corner_source_y >= 0 and
|
|
corner_source_x + overlap_x <= final_width and
|
|
corner_source_y + overlap_y <= final_height):
|
|
working_tensor[0, :overlap_y, :overlap_x, :] = final_tensor[0,
|
|
corner_source_y:corner_source_y + overlap_y,
|
|
corner_source_x:corner_source_x + overlap_x,
|
|
:]
|
|
outpaint_mask[0, :overlap_y, :overlap_x, :] = 0
|
|
|
|
# Get latent shape for the uniform grown canvas
|
|
with torch.no_grad():
|
|
latent_shape = vae.encode(working_tensor).shape
|
|
|
|
# Create empty latent tensor with grown shape
|
|
latent_image = torch.zeros(latent_shape, device=device)
|
|
|
|
# Apply ControlNet to conditioning using the grown canvas
|
|
conditioning = apply_controlnet_to_conditioning(
|
|
positive=pos_cond,
|
|
negative=neg_cond,
|
|
control_net=controlnet,
|
|
image=working_tensor,
|
|
strength=controlnet_strength,
|
|
start_percent=0.0,
|
|
end_percent=1.0,
|
|
vae=vae
|
|
)
|
|
|
|
# Generate noise for grown canvas size
|
|
noise = comfy.sample.prepare_noise(latent_image, current_seed, None)
|
|
|
|
# Sample with the grown latent
|
|
samples = comfy.sample.sample(
|
|
model,
|
|
noise,
|
|
steps,
|
|
cfg,
|
|
sampler_name,
|
|
scheduler,
|
|
conditioning[0],
|
|
conditioning[1],
|
|
latent_image
|
|
)
|
|
|
|
# Decode the full grown canvas
|
|
full_grown_tensor = vae.decode(samples)
|
|
|
|
# EXTRACT the tile_width x tile_height region from consistent position
|
|
# Always extract from overlap_x, overlap_y for uniform behavior
|
|
extract_x = overlap_x
|
|
extract_y = overlap_y
|
|
|
|
extracted_tile = full_grown_tensor[0:1,
|
|
extract_y:extract_y + tile_height,
|
|
extract_x:extract_x + tile_width,
|
|
:]
|
|
|
|
# Store both versions
|
|
individual_tensors.append(extracted_tile.clone()) # For blending
|
|
full_grown_tensors.append(full_grown_tensor.clone()) # For debugging
|
|
positions.append((final_pos_x, final_pos_y))
|
|
|
|
# Place the tile in the final composite at exact grid position
|
|
final_tensor[0,
|
|
final_pos_y:final_pos_y + tile_height,
|
|
final_pos_x:final_pos_x + tile_width,
|
|
:] = extracted_tile[0, :, :, :]
|
|
|
|
# Update progress
|
|
pbar.update(1)
|
|
|
|
# Apply Gaussian blending for smooth transitions
|
|
if individual_tensors:
|
|
final_tensor = gaussian_blend_tiles(
|
|
individual_tensors, # Use cropped tiles for blending
|
|
positions,
|
|
tile_width,
|
|
tile_height,
|
|
overlap_x,
|
|
overlap_y,
|
|
final_width,
|
|
final_height,
|
|
sigma=blend_sigma
|
|
)
|
|
# Return full grown tiles for debugging
|
|
tile_batch = torch.cat(full_grown_tensors, dim=0)
|
|
else:
|
|
final_tensor = torch.zeros((1, final_height, final_width, 3), dtype=torch.float32)
|
|
tile_batch = torch.zeros((1, gen_height, gen_width, 3), dtype=torch.float32, device=device)
|
|
|
|
return final_tensor, tile_batch
|
|
|
|
# This part is needed for ComfyUI to recognize the nodes
|
|
NODE_CLASS_MAPPINGS = {
|
|
"TiledImageGenerator": TiledImageGenerator,
|
|
}
|
|
|
|
# Add descriptions for the web UI
|
|
NODE_DISPLAY_NAME_MAPPINGS = {
|
|
"TiledImageGenerator": "Tiled Image Generator",
|
|
}
|