63 lines
2.4 KiB
Python
63 lines
2.4 KiB
Python
"""
|
|
XB-ToolBox FLUX多图编码节点
|
|
============================
|
|
套壳官方的 ResizeImagesByLongerEdge + VAEEncode + ReferenceLatent 串联。
|
|
"""
|
|
|
|
import node_helpers
|
|
import nodes
|
|
from comfy_extras.nodes_dataset import ResizeImagesByLongerEdgeNode
|
|
|
|
|
|
class XB_FluxMultiImageEncode:
|
|
"""FLUX多图编码 — 缩放→VAE编码→ReferenceLatent串联。
|
|
VAE和图片1为必需，图片2/3可选接入。
|
|
输出正面条件与负面条件（两个并行的ReferenceLatent链）。"""
|
|
|
|
@classmethod
|
|
def INPUT_TYPES(cls):
|
|
return {
|
|
"required": {
|
|
"positive": ("CONDITIONING",),
|
|
"negative": ("CONDITIONING",),
|
|
"vae": ("VAE",),
|
|
"image1": ("IMAGE",),
|
|
"longer_edge": ("INT", {"default": 1024, "min": 64, "max": 8192, "step": 64}),
|
|
},
|
|
"optional": {
|
|
"image2": ("IMAGE",),
|
|
"image3": ("IMAGE",),
|
|
},
|
|
}
|
|
|
|
RETURN_TYPES = ("CONDITIONING", "CONDITIONING")
|
|
RETURN_NAMES = ("正面条件", "负面条件")
|
|
FUNCTION = "encode"
|
|
CATEGORY = "XB_ToolBox/FLUX"
|
|
|
|
def encode(self, positive, negative, vae, image1, longer_edge, image2=None, image3=None):
|
|
# ── 收集图片，缩放后 VAE 编码 ──
|
|
latents = []
|
|
for img in [image1, image2, image3]:
|
|
if img is not None:
|
|
# 官方 ResizeImagesByLongerEdge（缩放图像-长边）
|
|
resized = ResizeImagesByLongerEdgeNode._process(img, longer_edge=longer_edge)
|
|
# 官方 VAEEncode
|
|
latent = nodes.VAEEncode().encode(vae, resized)[0]
|
|
latents.append(latent)
|
|
|
|
# ── 正面条件链（从输入的 positive 开始串联）──
|
|
pos_cond = positive
|
|
for lat in latents:
|
|
pos_cond = node_helpers.conditioning_set_values(
|
|
pos_cond, {"reference_latents": [lat["samples"]]}, append=True)
|
|
|
|
# ── 负面条件链（从输入的 negative 开始串联）──
|
|
neg_cond = negative
|
|
for lat in latents:
|
|
neg_cond = node_helpers.conditioning_set_values(
|
|
neg_cond, {"reference_latents": [lat["samples"]]}, append=True)
|
|
|
|
print(f"\033[92m[FLUX多图编码]\033[0m 图片数: {len(latents)} | 长边: {longer_edge}px")
|
|
return (pos_cond, neg_cond)
|