feat: 🚀 add seamless model hack

Inspired by the A111 hack and FlyingFireCo/tiled_ksampler
This commit is contained in:
melMass
2023-10-05 18:28:36 +02:00
parent d49b2578c2
commit 21acc87ff0
4 changed files with 1066 additions and 28 deletions
+905
View File
@@ -0,0 +1,905 @@
{
"last_node_id": 97,
"last_link_id": 179,
"nodes": [
{
"id": 6,
"type": "CLIPTextEncode",
"pos": [
-1165.8749246009997,
30
],
"size": [
422.84503173828125,
164.31304931640625
],
"flags": {},
"order": 4,
"mode": 0,
"inputs": [
{
"name": "clip",
"type": "CLIP",
"link": 3
}
],
"outputs": [
{
"name": "CONDITIONING",
"type": "CONDITIONING",
"links": [
4,
158
],
"slot_index": 0
}
],
"properties": {
"Node name for S&R": "CLIPTextEncode"
},
"widgets_values": [
"Closeup texture of rocks"
],
"color": "#432",
"bgcolor": "#653",
"shape": 1
},
{
"id": 7,
"type": "CLIPTextEncode",
"pos": [
-1175.8749246009997,
250
],
"size": [
425.27801513671875,
180.6060791015625
],
"flags": {},
"order": 5,
"mode": 0,
"inputs": [
{
"name": "clip",
"type": "CLIP",
"link": 5
}
],
"outputs": [
{
"name": "CONDITIONING",
"type": "CONDITIONING",
"links": [
6
],
"slot_index": 0
}
],
"properties": {
"Node name for S&R": "CLIPTextEncode"
},
"widgets_values": [
"((drawing, cartoon, painting, sketch, blur, depth of field, dof))"
],
"color": "#432",
"bgcolor": "#653",
"shape": 1
},
{
"id": 89,
"type": "Reroute",
"pos": [
350,
803
],
"size": [
75,
26
],
"flags": {},
"order": 9,
"mode": 0,
"inputs": [
{
"name": "",
"type": "*",
"link": 176
}
],
"outputs": [
{
"name": "",
"type": "IMAGE",
"links": [
167
]
}
],
"properties": {
"showOutputText": false,
"horizontal": false
}
},
{
"id": 4,
"type": "CheckpointLoaderSimple",
"pos": [
-1740,
236
],
"size": [
315,
98
],
"flags": {},
"order": 0,
"mode": 0,
"outputs": [
{
"name": "MODEL",
"type": "MODEL",
"links": [
170
],
"slot_index": 0
},
{
"name": "CLIP",
"type": "CLIP",
"links": [
3,
5
],
"slot_index": 1
},
{
"name": "VAE",
"type": "VAE",
"links": [],
"slot_index": 2
}
],
"properties": {
"Node name for S&R": "CheckpointLoaderSimple"
},
"widgets_values": [
"revAnimated_v122.safetensors"
],
"shape": 1
},
{
"id": 63,
"type": "SaveImage",
"pos": [
1315,
18
],
"size": [
539.2050170898438,
617.2159423828125
],
"flags": {},
"order": 13,
"mode": 0,
"inputs": [
{
"name": "images",
"type": "IMAGE",
"link": 115
}
],
"title": "Normal",
"properties": {},
"widgets_values": [
"Normal"
],
"shape": 1
},
{
"id": 67,
"type": "SaveImage",
"pos": [
2095,
22
],
"size": [
539.2050170898438,
617.2159423828125
],
"flags": {},
"order": 16,
"mode": 0,
"inputs": [
{
"name": "images",
"type": "IMAGE",
"link": 119
}
],
"title": "Curvature",
"properties": {},
"widgets_values": [
"Curvature"
],
"shape": 1
},
{
"id": 69,
"type": "SaveImage",
"pos": [
1560,
1290
],
"size": [
539.2050170898438,
617.2159423828125
],
"flags": {},
"order": 17,
"mode": 0,
"inputs": [
{
"name": "images",
"type": "IMAGE",
"link": 121
}
],
"title": "Depth",
"properties": {},
"widgets_values": [
"Height"
],
"shape": 1
},
{
"id": 91,
"type": "Model Patch Seamless (mtb)",
"pos": [
-1150,
-146
],
"size": [
430.8000183105469,
78
],
"flags": {},
"order": 3,
"mode": 0,
"inputs": [
{
"name": "model",
"type": "MODEL",
"link": 170
}
],
"outputs": [
{
"name": "Original Model (passthrough)",
"type": "MODEL",
"links": null,
"shape": 3
},
{
"name": "Patched Model",
"type": "MODEL",
"links": [
169
],
"shape": 3,
"slot_index": 1
}
],
"properties": {
"Node name for S&R": "Model Patch Seamless (mtb)"
},
"widgets_values": [
true
],
"color": "#232",
"bgcolor": "#353"
},
{
"id": 93,
"type": "PreviewImage",
"pos": [
1115,
-597
],
"size": [
451.3526306152344,
478.3444519042969
],
"flags": {},
"order": 12,
"mode": 0,
"inputs": [
{
"name": "images",
"type": "IMAGE",
"link": 179
}
],
"properties": {
"Node name for S&R": "PreviewImage"
}
},
{
"id": 43,
"type": "VAELoader",
"pos": [
-598.2757622278747,
577.3595309932109
],
"size": [
387.48089599609375,
70.60645294189453
],
"flags": {},
"order": 1,
"mode": 0,
"outputs": [
{
"name": "VAE",
"type": "VAE",
"links": [
174
],
"slot_index": 0
}
],
"properties": {
"Node name for S&R": "VAELoader"
},
"widgets_values": [
"vae-ft-mse-840000-ema-pruned.safetensors"
],
"shape": 1
},
{
"id": 97,
"type": "Image Tile Offset (mtb)",
"pos": [
617,
-598
],
"size": [
315,
58
],
"flags": {},
"order": 10,
"mode": 0,
"inputs": [
{
"name": "image",
"type": "IMAGE",
"link": 178
}
],
"outputs": [
{
"name": "IMAGE",
"type": "IMAGE",
"links": [
179
],
"shape": 3,
"slot_index": 0
}
],
"properties": {
"Node name for S&R": "Image Tile Offset (mtb)"
},
"widgets_values": [
2
],
"color": "#232",
"bgcolor": "#353"
},
{
"id": 96,
"type": "Vae Decode (mtb)",
"pos": [
-52,
40
],
"size": [
315,
126
],
"flags": {},
"order": 7,
"mode": 0,
"inputs": [
{
"name": "samples",
"type": "LATENT",
"link": 173
},
{
"name": "vae",
"type": "VAE",
"link": 174
}
],
"outputs": [
{
"name": "IMAGE",
"type": "IMAGE",
"links": [
175,
176,
178
],
"shape": 3,
"slot_index": 0
}
],
"properties": {
"Node name for S&R": "Vae Decode (mtb)"
},
"widgets_values": [
true,
false,
512
],
"color": "#232",
"bgcolor": "#353"
},
{
"id": 46,
"type": "SaveImage",
"pos": [
533,
25
],
"size": [
539.2050170898438,
617.2159423828125
],
"flags": {},
"order": 8,
"mode": 0,
"inputs": [
{
"name": "images",
"type": "IMAGE",
"link": 175
}
],
"title": "Albedo",
"properties": {},
"widgets_values": [
"Albedo"
],
"shape": 1
},
{
"id": 74,
"type": "EmptyLatentImage",
"pos": [
-1075.8749246009997,
480
],
"size": [
315,
106
],
"flags": {},
"order": 2,
"mode": 0,
"outputs": [
{
"name": "LATENT",
"type": "LATENT",
"links": [
132
],
"shape": 3,
"slot_index": 0
}
],
"properties": {
"Node name for S&R": "EmptyLatentImage"
},
"widgets_values": [
768,
768,
1
],
"color": "#323",
"bgcolor": "#535",
"shape": 1
},
{
"id": 62,
"type": "Deep Bump (mtb)",
"pos": [
727,
801
],
"size": [
315,
130
],
"flags": {},
"order": 11,
"mode": 0,
"inputs": [
{
"name": "image",
"type": "IMAGE",
"link": 167
}
],
"outputs": [
{
"name": "IMAGE",
"type": "IMAGE",
"links": [
115,
118,
122
],
"shape": 3,
"slot_index": 0
}
],
"properties": {
"Node name for S&R": "Deep Bump (mtb)"
},
"widgets_values": [
"Color to Normals",
"SMALL",
"SMALLEST",
true
],
"color": "#232",
"bgcolor": "#353",
"shape": 1
},
{
"id": 66,
"type": "Deep Bump (mtb)",
"pos": [
1626,
808
],
"size": [
315,
130
],
"flags": {},
"order": 14,
"mode": 0,
"inputs": [
{
"name": "image",
"type": "IMAGE",
"link": 118
}
],
"outputs": [
{
"name": "IMAGE",
"type": "IMAGE",
"links": [
119
],
"shape": 3,
"slot_index": 0
}
],
"properties": {
"Node name for S&R": "Deep Bump (mtb)"
},
"widgets_values": [
"Normals to Curvature",
"SMALL",
"SMALLEST",
true
],
"color": "#232",
"bgcolor": "#353",
"shape": 1
},
{
"id": 68,
"type": "Deep Bump (mtb)",
"pos": [
1185,
1288
],
"size": [
315,
130
],
"flags": {},
"order": 15,
"mode": 0,
"inputs": [
{
"name": "image",
"type": "IMAGE",
"link": 122
}
],
"outputs": [
{
"name": "IMAGE",
"type": "IMAGE",
"links": [
121
],
"shape": 3,
"slot_index": 0
}
],
"properties": {
"Node name for S&R": "Deep Bump (mtb)"
},
"widgets_values": [
"Normals to Height",
"SMALL",
"SMALLEST",
true
],
"color": "#232",
"bgcolor": "#353",
"shape": 1
},
{
"id": 3,
"type": "KSampler",
"pos": [
-518.2757622278748,
47.359530993211024
],
"size": [
315,
474
],
"flags": {},
"order": 6,
"mode": 0,
"inputs": [
{
"name": "model",
"type": "MODEL",
"link": 169
},
{
"name": "positive",
"type": "CONDITIONING",
"link": 4
},
{
"name": "negative",
"type": "CONDITIONING",
"link": 6
},
{
"name": "latent_image",
"type": "LATENT",
"link": 132
}
],
"outputs": [
{
"name": "LATENT",
"type": "LATENT",
"links": [
173
],
"slot_index": 0
}
],
"properties": {
"Node name for S&R": "KSampler"
},
"widgets_values": [
1001,
"fixed",
28,
8,
"dpmpp_2m",
"normal",
1
],
"color": "#222",
"bgcolor": "#000",
"shape": 1
}
],
"links": [
[
3,
4,
1,
6,
0,
"CLIP"
],
[
4,
6,
0,
3,
1,
"CONDITIONING"
],
[
5,
4,
1,
7,
0,
"CLIP"
],
[
6,
7,
0,
3,
2,
"CONDITIONING"
],
[
115,
62,
0,
63,
0,
"IMAGE"
],
[
118,
62,
0,
66,
0,
"IMAGE"
],
[
119,
66,
0,
67,
0,
"IMAGE"
],
[
121,
68,
0,
69,
0,
"IMAGE"
],
[
122,
62,
0,
68,
0,
"IMAGE"
],
[
132,
74,
0,
3,
3,
"LATENT"
],
[
158,
6,
0,
86,
0,
"*"
],
[
167,
89,
0,
62,
0,
"IMAGE"
],
[
169,
91,
1,
3,
0,
"MODEL"
],
[
170,
4,
0,
91,
0,
"MODEL"
],
[
173,
3,
0,
96,
0,
"LATENT"
],
[
174,
43,
0,
96,
1,
"VAE"
],
[
175,
96,
0,
46,
0,
"IMAGE"
],
[
176,
96,
0,
89,
0,
"*"
],
[
178,
96,
0,
97,
0,
"IMAGE"
],
[
179,
97,
0,
93,
0,
"IMAGE"
]
],
"groups": [
{
"title": "Seamless Diffusion",
"bounding": [
-1752,
-392,
1658,
1102
],
"color": "#3f789e",
"font_size": 76,
"locked": false
},
{
"title": "Seamless Check",
"bounding": [
421,
-795,
1374,
763
],
"color": "#3f789e",
"font_size": 76,
"locked": false
}
],
"config": {},
"extra": {},
"version": 0.4
}
+3 -11
View File
@@ -1,4 +1,3 @@
import pathlib
import tempfile
from pathlib import Path
@@ -8,15 +7,8 @@ import torch
from PIL import Image
from ..log import mklog
from ..utils import (
models_dir,
np2tensor,
pil2tensor,
tensor2pil,
tiles_infer,
tiles_merge,
tiles_split,
)
from ..utils import (models_dir, tensor2pil, tiles_infer, tiles_merge,
tiles_split)
# Disable MS telemetry
ort.disable_telemetry_events()
@@ -306,7 +298,7 @@ class DeepBump:
"LARGEST",
],
),
"normals_to_height_seamless": ("BOOLEAN", {"default": False}),
"normals_to_height_seamless": ("BOOLEAN", {"default": True}),
},
}
+68 -17
View File
@@ -1,17 +1,19 @@
import torch
from skimage.filters import gaussian
from skimage.util import compare_images
import itertools
import json
import math
import os
import cv2
import folder_paths
import numpy as np
import torch
import torch.nn.functional as F
from PIL import Image
from ..utils import tensor2pil, pil2tensor, tensor2np
import torch
import folder_paths
from PIL.PngImagePlugin import PngInfo
import json
import os
import math
from skimage.filters import gaussian
from skimage.util import compare_images
from ..utils import pil2tensor, tensor2np, tensor2pil
# try:
# from cv2.ximgproc import guidedFilter
@@ -177,7 +179,7 @@ class ColorCorrect:
return (image,)
class ImageCompare:
class ImageCompare_:
"""Compare two images and return a difference image"""
@classmethod
@@ -213,7 +215,7 @@ class ImageCompare:
import requests
class LoadImageFromUrl:
class LoadImageFromUrl_:
"""Load an image from the given URL"""
@classmethod
@@ -239,7 +241,7 @@ class LoadImageFromUrl:
return (pil2tensor(image),)
class Blur:
class Blur_:
"""Blur an image using a Gaussian filter."""
@classmethod
@@ -501,7 +503,7 @@ class ImageResizeFactor:
return (resized_image,)
class SaveImageGrid:
class SaveImageGrid_:
"""Save all the images in the input batch as a grid of images."""
def __init__(self):
@@ -603,15 +605,64 @@ class SaveImageGrid:
return {"ui": {"images": results}}
class ImageTileOffset:
"""Mimics an old photoshop technique to check for seamless textures"""
@classmethod
def INPUT_TYPES(cls):
return {
"required": {
"image": ("IMAGE",),
"tiles": ("INT", {"default": 2}),
}
}
CATEGORY = "mtb/generate"
RETURN_TYPES = ("IMAGE",)
FUNCTION = "tile_image"
def tile_image(self, image: torch.Tensor, tiles: int = 2):
if tiles < 1:
raise ValueError("The number of tiles must be at least 1.")
batch_size, height, width, channels = image.shape
tile_height = height // tiles
tile_width = width // tiles
output_image = torch.zeros_like(image)
for i, j in itertools.product(range(tiles), range(tiles)):
start_h = i * tile_height
end_h = start_h + tile_height
start_w = j * tile_width
end_w = start_w + tile_width
tile = image[:, start_h:end_h, start_w:end_w, :]
output_start_h = (i + 1) % tiles * tile_height
output_start_w = (j + 1) % tiles * tile_width
output_end_h = output_start_h + tile_height
output_end_w = output_start_w + tile_width
output_image[
:, output_start_h:output_end_h, output_start_w:output_end_w, :
] = tile
return (output_image,)
__nodes__ = [
ColorCorrect,
ImageCompare,
Blur,
ImageCompare_,
ImageTileOffset,
Blur_,
# DeglazeImage,
MaskToImage,
ColoredImage,
ImagePremultiply,
ImageResizeFactor,
SaveImageGrid,
LoadImageFromUrl,
SaveImageGrid_,
LoadImageFromUrl_,
]
+90
View File
@@ -0,0 +1,90 @@
import copy
import torch
class VaeDecode_:
"""Wrapper for the 2 core decoders but also adding the sd seamless hack, taken from: FlyingFireCo/tiled_ksampler"""
@classmethod
def INPUT_TYPES(cls):
return {
"required": {
"samples": ("LATENT",),
"vae": ("VAE",),
"seamless_model": ("BOOLEAN", {"default": False}),
"use_tiling_decoder": ("BOOLEAN", {"default": True}),
"tile_size": (
"INT",
{"default": 512, "min": 320, "max": 4096, "step": 64},
),
}
}
RETURN_TYPES = ("IMAGE",)
FUNCTION = "decode"
CATEGORY = "mtb/decode"
def decode(
self, vae, samples, seamless_model, use_tiling_decoder=True, tile_size=512
):
if seamless_model:
for layer in [
layer
for layer in vae.first_stage_model.modules()
if isinstance(layer, torch.nn.Conv2d)
]:
layer.padding_mode = "circular"
if use_tiling_decoder:
return (
vae.decode_tiled(
samples["samples"],
tile_x=tile_size // 8,
tile_y=tile_size // 8,
),
)
else:
return (vae.decode(samples["samples"]),)
class ModelPatchSeamless:
@classmethod
def INPUT_TYPES(cls):
return {
"required": {
"model": ("MODEL",),
"tiling": (
"BOOLEAN",
{"default": True},
), # kept for testing not sure why it should be false
}
}
RETURN_TYPES = ("MODEL", "MODEL")
RETURN_NAMES = (
"Original Model (passthrough)",
"Patched Model",
)
FUNCTION = "hack"
CATEGORY = "mtb/textures"
def apply_circular(self, model, enable):
for layer in [
layer for layer in model.modules() if isinstance(layer, torch.nn.Conv2d)
]:
layer.padding_mode = "circular" if enable else "zeros"
return model
def hack(
self,
model,
tiling,
):
hacked_model = copy.deepcopy(model)
self.apply_circular(hacked_model.model, tiling)
return (model, hacked_model)
__nodes__ = [ModelPatchSeamless, VaeDecode_]