🚀 Flux-Dalle-Mix-LoRA [實驗版本]
Flux-Dalle-Mix-LoRA是一個用於圖像生成的模型,它結合了Dalle和Flux的特性,能夠生成逼真且富有創意的圖像。不過該模型仍處於訓練階段,當前並非最終版本,可能存在瑕疵,某些情況下表現欠佳。
立即試用
⚠️ 重要提示
該模型仍處於訓練階段,這不是最終版本,可能包含瑕疵,在某些情況下性能可能不佳。
✨ 主要特性
- 多風格圖像生成:支持生成攝影、卡通、3D渲染等多種風格的圖像。
- 高逼真度:能夠生成具有高度真實感的圖像,如人物肖像、動物特寫等。
- 參數可調節:通過調整圖像生成參數,可以滿足不同的創作需求。
📦 安裝指南
import torch
from pipelines import DiffusionPipeline
base_model = "black-forest-labs/FLUX.1-dev"
pipe = DiffusionPipeline.from_pretrained(base_model, torch_dtype=torch.bfloat16)
lora_repo = "prithivMLmods/Flux-Dalle-Mix-LoRA"
trigger_word = "dalle-mix"
pipe.load_lora_weights(lora_repo)
device = torch.device("cuda")
pipe.to(device)
💻 使用示例
基礎用法
import torch
from pipelines import DiffusionPipeline
base_model = "black-forest-labs/FLUX.1-dev"
pipe = DiffusionPipeline.from_pretrained(base_model, torch_dtype=torch.bfloat16)
lora_repo = "prithivMLmods/Flux-Dalle-Mix-LoRA"
trigger_word = "dalle-mix"
pipe.load_lora_weights(lora_repo)
device = torch.device("cuda")
pipe.to(device)
prompt = "dalle-mix, photography, portrait of contrast, profile silhouette of a woman, vibrant hot pink backdrop, visualize using a camera setup that mimics a large aperture, focusing solely on the silhouettes edge, while a low ISO maintains the richness of color without grain, photorealistic, UHD --ar 9:16 --chaos 1.7 --style raw"
image = pipe(prompt).images[0]
image.save("output.png")
高級用法
在高級用法中,你可以根據需要調整更多的參數,如LR Scheduler
、Optimizer
等,以獲得更好的圖像生成效果。以下是一個示例:
parameters = {
"LR Scheduler": "constant",
"Noise Offset": 0.03,
"Optimizer": "AdamW",
"Multires Noise Discount": 0.1,
"Network Dim": 64,
"Network Alpha": 32,
"Epoch": 15,
"Save Every N Epochs": 1,
"Multires Noise Iterations": 10,
"Repeat & Steps": "25 & 3700"
}
for param, value in parameters.items():
setattr(pipe, param, value)
prompt = "dalle-mix, Ultra-detailed close-up of a pandas face, styled like a Pixar or DreamWorks character, highlight intricately textured fur for depth, the panda sports a pirate eyepatch, adding a roguish charm, he has a furrowed brow and a smirking mouth, showcasing a cunning and playful demeanor, emphasize crisp details in the facial features, particularly around the expressive eyes and smirking mouth, to bring out the character’s personality, 3D render, --v 6 --style raw --stylize 250 --ar 4:5"
image = pipe(prompt).images[0]
image.save("advanced_output.png")
📚 詳細文檔
模型描述
prithivMLmods/Flux-Dalle-Mix-LoRA
屬性 |
詳情 |
模型類型 |
LoRA(Low-Rank Adaptation) |
基礎模型 |
black-forest-labs/FLUX.1-dev |
實例提示詞 |
dalle-mix |
許可證 |
creativeml-openrail-m |
新版本 |
strangerzonehf/Flux-Midjourney-Mix-LoRA |
圖像生成參數
參數 |
值 |
參數 |
值 |
LR Scheduler |
constant |
Noise Offset |
0.03 |
Optimizer |
AdamW |
Multires Noise Discount |
0.1 |
Network Dim |
64 |
Multires Noise Iterations |
10 |
Network Alpha |
32 |
Repeat & Steps |
25 & 3700 |
Epoch |
15 |
Save Every N Epochs |
1 |
標註信息
標註使用florence2-en(自然語言 & 英語)
訓練數據
總共使用44張高分辨率圖像進行訓練
最佳尺寸
- 768 x 1024(最佳)
- 1024 x 1024(默認)
觸發詞
⚠️ 重要提示
你應該使用dalle-mix
來觸發圖像生成。
示例圖像
示例1

描述 |
dalle-mix, Ultra-detailed close-up of a pandas face, styled like a Pixar or DreamWorks character, highlight intricately textured fur for depth, the panda sports a pirate eyepatch, adding a roguish charm, he has a furrowed brow and a smirking mouth, showcasing a cunning and playful demeanor, emphasize crisp details in the facial features, particularly around the expressive eyes and smirking mouth, to bring out the character’s personality, 3D render, --v 6 --style raw --stylize 250 --ar 4:5 |
示例2

描述 |
dalle-mix, Caricatures pencil drawing, cute woman, mid-20's, color pencils, in the style of Pixar, hilariously funny, exaggerated features, white background --chaos 50 --ar 1:2 --stylize 750 |
📄 許可證
本模型使用creativeml-openrail-m
許可證。
📥 下載模型
該模型的權重以Safetensors格式提供。你可以在文件與版本選項卡中下載。