From 693919e787885be15e74938470c3d5d9ed341768 Mon Sep 17 00:00:00 2001 From: Yousef Rafat <81116377+yousef-rafat@users.noreply.github.com> Date: Wed, 29 Apr 2026 13:09:44 +0300 Subject: [PATCH] moving to nodes_images --- comfy_extras/nodes_images.py | 280 ++++++++++++++++++++++++++ comfy_extras/nodes_save_advanced.py | 297 ---------------------------- 2 files changed, 280 insertions(+), 297 deletions(-) delete mode 100644 comfy_extras/nodes_save_advanced.py diff --git a/comfy_extras/nodes_images.py b/comfy_extras/nodes_images.py index a77f0641f..5c32305db 100644 --- a/comfy_extras/nodes_images.py +++ b/comfy_extras/nodes_images.py @@ -4,15 +4,23 @@ import nodes import folder_paths import json +import av import os import re import math import torch +import struct +import zlib +import tempfile +import logging import comfy.utils +import numpy as np +from fractions import Fraction from server import PromptServer from comfy_api.latest import ComfyExtension, IO, UI from typing_extensions import override +from comfy.cli_args import args SVG = IO.SVG.Type # TODO: temporary solution for backward compatibility, will be removed later. @@ -823,6 +831,277 @@ class ImageMergeTileList(IO.ComfyNode): return IO.NodeOutput(merged_image) +def create_png_chunk(chunk_type: bytes, data: bytes) -> bytes: + """Creates a valid PNG chunk with Length, Type, Data, and CRC32.""" + chunk = struct.pack('>I', len(data)) + chunk_type + data + crc = zlib.crc32(chunk_type + data) & 0xffffffff + return chunk + struct.pack('>I', crc) + +def inject_comfy_metadata_png(png_bytes, prompt=None, extra_pnginfo=None): + # IEND chunk is the last 12 bytes of png files + content = png_bytes[:-12] + iend = png_bytes[-12:] + + metadata_chunks = b"" + + if prompt is not None: + payload = b'prompt\x00' + json.dumps(prompt).encode('utf-8') + metadata_chunks += create_png_chunk(b'tEXt', payload) + + if extra_pnginfo is not None: + for k, v in extra_pnginfo.items(): + payload = k.encode('utf-8') + b'\x00' + json.dumps(v).encode('utf-8') + metadata_chunks += create_png_chunk(b'tEXt', payload) + + return content + metadata_chunks + iend + +def inject_comfy_metadata_exr(exr_bytes: bytes, prompt, extra_pnginfo) -> bytes: + # skip magic and version + idx = 8 + + # parse through existing attributes to find the end of the header + while True: + name_start = idx + while exr_bytes[idx] != 0: + idx += 1 + name = exr_bytes[name_start:idx] + idx += 1 + + # empty name means we hit the header terminator + if len(name) == 0: + break + + # skip attribute type string + while exr_bytes[idx] != 0: + idx += 1 + idx += 1 + + # read attribute size and skip the value + attr_size = struct.unpack(' bytes: + metadata = {} + if prompt is not None: + metadata["prompt"] = prompt + if extra_pnginfo is not None: + for k, v in extra_pnginfo.items(): + metadata[k] = v + + payload = json.dumps(metadata).encode('utf-8') + + # 16-byte uuid required by isobmff spec + # 'comfyui_workflow' is exactly 16 bytes long! + comfy_uuid = b'comfyui_workflow' + + # box size: 4 (size) + 4 (type) + 16 (uuid) + payload length + box_size = 4 + 4 + 16 + len(payload) + uuid_box = struct.pack('>I', box_size) + b'uuid' + comfy_uuid + payload + + # isobmff allows top-level boxes at the end of the file. + return avif_bytes + uuid_box + +class SaveImageAdvanced(IO.ComfyNode): + @classmethod + def define_schema(cls): + return IO.Schema( + node_id="SaveImageAdvanced", + category="image/advanced_io", + output_node=True, + inputs=[ + IO.Image.Input("images"), + IO.String.Input("filename_prefix", default="ComfyUI"), + IO.Combo.Input("file_format", options=["png", "exr", "avif"], default="png"), + IO.Combo.Input("bit_depth", options=["8-bit", "16-bit", "32-bit"], default="8-bit"), + IO.Boolean.Input("embed_workflow", default=True), + IO.Hidden.Input("prompt", type="PROMPT"), + IO.Hidden.Input("extra_pnginfo", type="EXTRA_PNGINFO"), + ], + outputs=[] + ) + + @classmethod + def execute(cls, images, filename_prefix="ComfyUI", file_format="png", bit_depth="8-bit", + embed_workflow=True, prompt=None, extra_pnginfo=None) -> IO.NodeOutput: + + output_dir = folder_paths.get_output_directory() + + full_output_folder, filename, counter, subfolder, filename_prefix = \ + folder_paths.get_save_image_path(filename_prefix, output_dir, images[0].shape[1], images[0].shape[0]) + + results = list() + + for batch_number, image in enumerate(images): + img_tensor = image.clone() + + height, width, num_channels = img_tensor.shape + has_alpha = (num_channels == 4) + + # file pathing + filename_with_batch_num = filename.replace("%batch_num%", str(batch_number)) + file = f"{filename_with_batch_num}_{counter:05}.{file_format}" + file_path = os.path.join(full_output_folder, file) + + if file_format in ["png", "exr", "avif"]: + + # safe bit downcasting + if (file_format == "png" or file_format == "avif") and bit_depth == "32-bit": + bit_depth = "16-bit" + if file_format == "exr" and bit_depth == "8-bit": + bit_depth = "16-bit" + + if bit_depth == "32-bit": + img_np = img_tensor.cpu().numpy().astype(np.float32) + av_fmt = 'gbrapf32le' if has_alpha else 'gbrpf32le' + elif bit_depth == "16-bit": + if file_format == "exr": + # default pyav build doesn't come with a codec for float16 exr format + img_np = img_tensor.cpu().numpy().astype(np.float32) + av_fmt = 'gbrapf32le' if has_alpha else 'gbrpf32le' + else: + img_np = (img_tensor * 65535.0).clamp(0, 65535).to(torch.int32).cpu().numpy().astype(np.uint16) + av_fmt = 'rgba64le' if has_alpha else 'rgb48le' + else: + img_np = (img_tensor * 255.0).clamp(0, 255).to(torch.int32).cpu().numpy().astype(np.uint8) + av_fmt = 'rgba' if has_alpha else 'rgb24' + + fd, tmp_path = tempfile.mkstemp(suffix=f".{file_format}") + os.close(fd) + container_format = "image2" if file_format in ["png", "exr"] else "avif" + container = av.open(tmp_path, mode='w', format=container_format) + + if file_format == "exr": + stream = container.add_stream('exr', rate=1) + stream.pix_fmt = av_fmt + + elif file_format == "avif": + try: + stream = container.add_stream('libsvtav1', rate=1) + except Exception: + stream = container.add_stream('av1', rate=1) + + stream.time_base = Fraction(1, 1) + + if bit_depth in ["16-bit", "32-bit"]: + stream.pix_fmt = 'yuv420p10le' + else: + stream.pix_fmt = 'yuv420p' + + stream.codec_context.color_range = 2 + stream.codec_context.colorspace = 1 + stream.codec_context.color_primaries = 1 + stream.codec_context.color_trc = 1 + + stream.options = { + 'preset': '10', + 'svtav1-params': 'rc=0:qp=20:color-range=1:color-matrix=1:enable-overlays=1', + 'g': '1' + } + + elif file_format == "png": + stream = container.add_stream('png', rate=1) + if bit_depth == "16-bit": + stream.pix_fmt = 'rgba64be' if has_alpha else 'rgb48be' + else: + stream.pix_fmt = av_fmt + + stream.width = width + stream.height = height + stream.time_base = Fraction(1, 1) + + is_planar = av_fmt.startswith('gbrp') or 'p' in av_fmt.split('rgba')[-1] + if is_planar: + if av_fmt.startswith('gbrp'): + img_np = img_np[:, :, [1, 2, 0, 3]] if has_alpha else img_np[:, :, [1, 2, 0]] + img_np = img_np.transpose(2, 0, 1) + + try: + frame = av.VideoFrame.from_ndarray(img_np, format=av_fmt) + except ValueError: + logging.warning("[WARNING] Current FFMPEG Binary can't save natively. Fallbacking.") + img_np = (img_tensor * 65535.0).clamp(0, 65535).to(torch.int32).cpu().numpy().astype(np.uint16) + av_fmt = 'rgba64le' if has_alpha else 'rgb48le' + frame = av.VideoFrame.from_ndarray(img_np, format=av_fmt) + + # reformat for both avif and exr to ensure correct internal conversion + if file_format in ["avif", "exr"] or (file_format == "png" and bit_depth == "16-bit"): + reformat_kwargs = {"format": stream.pix_fmt} + if file_format == "avif": + reformat_kwargs.update({ + "src_colorspace": 1, "dst_colorspace": 1, + "src_color_range": 2, "dst_color_range": 2 + }) + frame = frame.reformat(**reformat_kwargs) + frame.pts = 0 + frame.time_base = stream.time_base + if file_format == "avif": + frame.color_range = 2 + frame.colorspace = 1 + + for packet in stream.encode(frame): + container.mux(packet) + for packet in stream.encode(): + container.mux(packet) + + container.close() + + with open(tmp_path, "rb") as f: + final_bytes = f.read() + os.remove(tmp_path) + + if embed_workflow and not args.disable_metadata: + if file_format == "png": + final_bytes = inject_comfy_metadata_png(final_bytes, prompt, extra_pnginfo) + elif file_format == "exr": + final_bytes = inject_comfy_metadata_exr(final_bytes, prompt, extra_pnginfo) + else: + final_bytes = inject_comfy_metadata_avif(final_bytes, prompt, extra_pnginfo) + + with open(file_path, "wb") as f: + f.write(final_bytes) + + results.append({ + "filename": file, + "subfolder": subfolder, + "type": "output" + }) + counter += 1 + + return IO.NodeOutput(ui={"images": results}) + class ImagesExtension(ComfyExtension): @override async def get_node_list(self) -> list[type[IO.ComfyNode]]: @@ -844,6 +1123,7 @@ class ImagesExtension(ComfyExtension): ImageScaleToMaxDimension, SplitImageToTileList, ImageMergeTileList, + SaveImageAdvanced ] diff --git a/comfy_extras/nodes_save_advanced.py b/comfy_extras/nodes_save_advanced.py deleted file mode 100644 index 91ac68edc..000000000 --- a/comfy_extras/nodes_save_advanced.py +++ /dev/null @@ -1,297 +0,0 @@ -import av -import os -import json -import torch -import struct -import zlib -import logging -import numpy as np -import tempfile -import folder_paths -from comfy_api.latest import IO -from typing_extensions import override -from comfy_api.latest import ComfyExtension -from comfy.cli_args import args -from fractions import Fraction - -def create_png_chunk(chunk_type: bytes, data: bytes) -> bytes: - """Creates a valid PNG chunk with Length, Type, Data, and CRC32.""" - chunk = struct.pack('>I', len(data)) + chunk_type + data - crc = zlib.crc32(chunk_type + data) & 0xffffffff - return chunk + struct.pack('>I', crc) - -def inject_comfy_metadata_png(png_bytes, prompt=None, extra_pnginfo=None): - # IEND chunk is the last 12 bytes of png files - content = png_bytes[:-12] - iend = png_bytes[-12:] - - metadata_chunks = b"" - - if prompt is not None: - payload = b'prompt\x00' + json.dumps(prompt).encode('utf-8') - metadata_chunks += create_png_chunk(b'tEXt', payload) - - if extra_pnginfo is not None: - for k, v in extra_pnginfo.items(): - payload = k.encode('utf-8') + b'\x00' + json.dumps(v).encode('utf-8') - metadata_chunks += create_png_chunk(b'tEXt', payload) - - return content + metadata_chunks + iend - -def inject_comfy_metadata_exr(exr_bytes: bytes, prompt, extra_pnginfo) -> bytes: - # skip magic and version - idx = 8 - - # parse through existing attributes to find the end of the header - while True: - name_start = idx - while exr_bytes[idx] != 0: - idx += 1 - name = exr_bytes[name_start:idx] - idx += 1 - - # empty name means we hit the header terminator - if len(name) == 0: - break - - # skip attribute type string - while exr_bytes[idx] != 0: - idx += 1 - idx += 1 - - # read attribute size and skip the value - attr_size = struct.unpack(' bytes: - metadata = {} - if prompt is not None: - metadata["prompt"] = prompt - if extra_pnginfo is not None: - for k, v in extra_pnginfo.items(): - metadata[k] = v - - payload = json.dumps(metadata).encode('utf-8') - - # 16-byte uuid required by isobmff spec - # 'comfyui_workflow' is exactly 16 bytes long! - comfy_uuid = b'comfyui_workflow' - - # box size: 4 (size) + 4 (type) + 16 (uuid) + payload length - box_size = 4 + 4 + 16 + len(payload) - uuid_box = struct.pack('>I', box_size) + b'uuid' + comfy_uuid + payload - - # isobmff allows top-level boxes at the end of the file. - return avif_bytes + uuid_box - -class SaveImageAdvanced(IO.ComfyNode): - @classmethod - def define_schema(cls): - return IO.Schema( - node_id="SaveImageAdvanced", - category="image/advanced_io", - output_node=True, - inputs=[ - IO.Image.Input("images"), - IO.String.Input("filename_prefix", default="ComfyUI"), - IO.Combo.Input("file_format", options=["png", "exr", "avif"], default="png"), - IO.Combo.Input("bit_depth", options=["8-bit", "16-bit", "32-bit"], default="8-bit"), - IO.Boolean.Input("embed_workflow", default=True), - IO.Hidden.Input("prompt", type="PROMPT"), - IO.Hidden.Input("extra_pnginfo", type="EXTRA_PNGINFO"), - ], - outputs=[] - ) - - @classmethod - def execute(cls, images, filename_prefix="ComfyUI", file_format="png", bit_depth="8-bit", - embed_workflow=True, prompt=None, extra_pnginfo=None) -> IO.NodeOutput: - - output_dir = folder_paths.get_output_directory() - - full_output_folder, filename, counter, subfolder, filename_prefix = \ - folder_paths.get_save_image_path(filename_prefix, output_dir, images[0].shape[1], images[0].shape[0]) - - results = list() - - for batch_number, image in enumerate(images): - img_tensor = image.clone() - - height, width, num_channels = img_tensor.shape - has_alpha = (num_channels == 4) - - # file pathing - filename_with_batch_num = filename.replace("%batch_num%", str(batch_number)) - file = f"{filename_with_batch_num}_{counter:05}.{file_format}" - file_path = os.path.join(full_output_folder, file) - - if file_format in ["png", "exr", "avif"]: - - # safe bit downcasting - if (file_format == "png" or file_format == "avif") and bit_depth == "32-bit": - bit_depth = "16-bit" - if file_format == "exr" and bit_depth == "8-bit": - bit_depth = "16-bit" - - if bit_depth == "32-bit": - img_np = img_tensor.cpu().numpy().astype(np.float32) - av_fmt = 'gbrapf32le' if has_alpha else 'gbrpf32le' - elif bit_depth == "16-bit": - if file_format == "exr": - # default pyav build doesn't come with a codec for float16 exr format - img_np = img_tensor.cpu().numpy().astype(np.float32) - av_fmt = 'gbrapf32le' if has_alpha else 'gbrpf32le' - else: - img_np = (img_tensor * 65535.0).clamp(0, 65535).to(torch.int32).cpu().numpy().astype(np.uint16) - av_fmt = 'rgba64le' if has_alpha else 'rgb48le' - else: - img_np = (img_tensor * 255.0).clamp(0, 255).to(torch.int32).cpu().numpy().astype(np.uint8) - av_fmt = 'rgba' if has_alpha else 'rgb24' - - fd, tmp_path = tempfile.mkstemp(suffix=f".{file_format}") - os.close(fd) - container_format = "image2" if file_format in ["png", "exr"] else "avif" - container = av.open(tmp_path, mode='w', format=container_format) - - if file_format == "exr": - stream = container.add_stream('exr', rate=1) - stream.pix_fmt = av_fmt - - elif file_format == "avif": - try: - stream = container.add_stream('libsvtav1', rate=1) - except Exception: - stream = container.add_stream('av1', rate=1) - - stream.time_base = Fraction(1, 1) - - if bit_depth in ["16-bit", "32-bit"]: - stream.pix_fmt = 'yuv420p10le' - else: - stream.pix_fmt = 'yuv420p' - - stream.codec_context.color_range = 2 - stream.codec_context.colorspace = 1 - stream.codec_context.color_primaries = 1 - stream.codec_context.color_trc = 1 - - stream.options = { - 'preset': '10', - 'svtav1-params': 'rc=0:qp=20:color-range=1:color-matrix=1:enable-overlays=1', - 'g': '1' - } - - elif file_format == "png": - stream = container.add_stream('png', rate=1) - if bit_depth == "16-bit": - stream.pix_fmt = 'rgba64be' if has_alpha else 'rgb48be' - else: - stream.pix_fmt = av_fmt - - stream.width = width - stream.height = height - stream.time_base = Fraction(1, 1) - - is_planar = av_fmt.startswith('gbrp') or 'p' in av_fmt.split('rgba')[-1] - if is_planar: - if av_fmt.startswith('gbrp'): - img_np = img_np[:, :, [1, 2, 0, 3]] if has_alpha else img_np[:, :, [1, 2, 0]] - img_np = img_np.transpose(2, 0, 1) - - try: - frame = av.VideoFrame.from_ndarray(img_np, format=av_fmt) - except ValueError: - logging.warning("[WARNING] Current FFMPEG Binary can't save natively. Fallbacking.") - img_np = (img_tensor * 65535.0).clamp(0, 65535).to(torch.int32).cpu().numpy().astype(np.uint16) - av_fmt = 'rgba64le' if has_alpha else 'rgb48le' - frame = av.VideoFrame.from_ndarray(img_np, format=av_fmt) - - # reformat for both avif and exr to ensure correct internal conversion - if file_format in ["avif", "exr"] or (file_format == "png" and bit_depth == "16-bit"): - reformat_kwargs = {"format": stream.pix_fmt} - if file_format == "avif": - reformat_kwargs.update({ - "src_colorspace": 1, "dst_colorspace": 1, - "src_color_range": 2, "dst_color_range": 2 - }) - frame = frame.reformat(**reformat_kwargs) - frame.pts = 0 - frame.time_base = stream.time_base - if file_format == "avif": - frame.color_range = 2 - frame.colorspace = 1 - - for packet in stream.encode(frame): - container.mux(packet) - for packet in stream.encode(): - container.mux(packet) - - container.close() - - with open(tmp_path, "rb") as f: - final_bytes = f.read() - os.remove(tmp_path) - - if embed_workflow and not args.disable_metadata: - if file_format == "png": - final_bytes = inject_comfy_metadata_png(final_bytes, prompt, extra_pnginfo) - elif file_format == "exr": - final_bytes = inject_comfy_metadata_exr(final_bytes, prompt, extra_pnginfo) - else: - final_bytes = inject_comfy_metadata_avif(final_bytes, prompt, extra_pnginfo) - - with open(file_path, "wb") as f: - f.write(final_bytes) - - results.append({ - "filename": file, - "subfolder": subfolder, - "type": "output" - }) - counter += 1 - - return IO.NodeOutput(ui={"images": results}) - -class AdvancedImageSave(ComfyExtension): - @override - async def get_node_list(self) -> list[type[IO.ComfyNode]]: - return [ - SaveImageAdvanced, - ] - - -async def comfy_entrypoint() -> AdvancedImageSave: - return AdvancedImageSave()