基于参考图室内装修_多选+放大 V2Pro丨Interior decoration based on reference pictures_multiple selection + upscale

5.0

1 reviews
36
5.6K
1.6K
20
Description

室内装修风格可视化 V2改进版

V2优化

  • 对空间的控制比前版本更准确
  • 增加了卧室适配,在上传图片下方,选择1适配客厅,选择2适配卧室
  • 室内家居效果更真实
  • 更便捷,(云版本专属)点击生成后自动完成两张效果图的生成和放大,无需人工干预

云版本链接:

https://runninghub.cn/#/post/1836286479432663042/?utm_source=kol01-RH009


V1介绍
这是一个不需要手动输入提示词,就将毛坯房间转化成指定装修风格的工作流

比如上传一张客厅毛坯图,再上传一张你想要参考的客厅风格图,工作流会执行两个阶段任务:

第一阶段,根据参考图生成四张粗略的装修风格图;

第二阶段,工作流会对进一步细化你选择的图,最后得到四张高清放大的效果图。


如果你是甲方业主,你可以通过这个工作流将你的想法可视化,交给设计师照完成细化设计,避免反复沟通耗时耗力;

如果你是乙方设计师,你可以通过这个工作流将甲方意向可视化,帮助TA完成决策,拿下订单,然后再回归细化设计。


如果好用请给我反馈,会持续迭代这个工作流

Enjoy!

-----------------------------------------------------------------------------------------------------

Visualization of interior decoration style V2 improved version

V2 optimization

  • Control of space is more accurate than the previous version
  • Added bedroom adaptation. Under the uploaded picture, select 1 to adapt to the living room and select 2 to adapt to the bedroom
  • Indoor home effect is more realistic
  • More convenient. (Cloud version only) After clicking generate, the generation and enlargement of two renderings are automatically completed without manual intervention
  • Cloud version avaliable on link:

https://runninghub.cn/#/post/1836286479432663042/?utm_source=kol01-RH009


V1 introduction

This is a workflow that converts a rough room into a specified decoration style without manually entering prompt words

For example, upload a rough living room picture, and then upload a living room style picture you want to refer to. The workflow will perform two-stage tasks:

In the first stage, four rough decoration style pictures are generated based on the reference picture;

In the second stage, the workflow will further refine the picture you selected, and finally get four high-definition enlarged renderings.

If you are the owner of Party A, you can use this workflow to visualize your ideas and hand them over to the designer to complete the detailed design to avoid time-consuming and labor-intensive repeated communication;

If you are a designer of Party B, you can use this workflow to visualize Party A's intentions, help TA make decisions, get orders, and then return to detailed design.

If it works well, please give me feedback. I will continue to iterate this workflow.

Enjoy!

Node Diagram
Discussion
u
umar khan6 months ago

it gives this error in downloadandrunflorence2

ImportError: cannot import name 'AutoConfig' from 'transformers'

J
Jay Lin6 months ago

check this repo for Florence2 installation issues:

https://github.com/kijai/ComfyUI-Florence2

u
umar khan6 months ago

tried reinstalling and updating transformers too,but still florence 2 gives error, workflow works great after removing florence 2 nodes


J
Jay Lin6 months ago

true, anything that gives captions of reference image works fine

👍1
祈烨6 months ago

我要怎么获得flux_室内现代风格.safetensors这个模型呢,找不到欸,也没有下载地址呢

👍2
t
tong liu6 months ago

jie,混元官方的深度模型一直加载出错,是怎么回事报错显示尺寸不对

h
harry chen5 months ago

Thanks for your workflow.

Where can I get HunyuanDiT_Controlnet_lineart.pt ?

J
Jay Lin5 months ago

https://huggingface.co/TTPlanet/HunyuanDiT_Controlnet_lineart

👍1
h
harry chen5 months ago

pytorch_model_lineart_ema_module_only.pt

我可以使用這個模型嗎?

Error occurred when executing LineArtPreprocessor: Input type (float) and bias type (c10::Half) should be the same

我收到這個錯誤

所以我認為 LINEART 模型有問題,你能教我嗎?

Thank you

(Edited)
J
Jay Lin5 months ago

https://huggingface.co/TTPlanet/HunyuanDiT_Controlnet_lineart/tree/main

是这里的模型没错,请确认模型正确加载

(Edited)
h
harry chen5 months ago

謝謝!

CHARLESa month ago

大佬,florence 2节点  那个flash attention 2 这个插件有必要装吗?好像挺复杂的。我该成sdpa跑有区别吗?但是出现错误:感谢回复!

LineArt艺术线预处理器

Input type (float) and bias type (struct c10::Half) should be the same

J
Jay Lina month ago

这个报错是controlnet模型精度不同的问题,跟florence2没关系,flash attention 2和sdpa两个都可以

CHARLESa month ago

好的。感谢回答。那可能是我用的大模型不一样有关,hunyuan的大模型还没下载下来,用的其他的通用模型

CHARLESa month ago

大佬。混元的这个大模型是不是生图的效果不是很好,第一个采样器出来的图都是有点糊的

Author

22
53.3K
1.2K
214.2K

Resources (3)

    卧室毛坯.jpg (1.3 MB)
    98ca66087f852783663b73c92e9496c7.jpg (36.8 kB)
    6a5c721a3adacd8ac5fca83ec53f4833.jpg (37 kB)

Reviews

h

harry chen

3 months ago

Versions (1)

  • - latest (6 months ago)

Primitive Nodes (52)

Anything Everywhere3 (1)

CLIPTextEncodeHunyuanDiT (2)

ControlNetApplySD3 (2)

DepthAnything_V2 (1)

DownloadAndLoadDepthAnythingV2Model (1)

DownloadAndLoadFlorence2Model (1)

Fast Groups Muter (rgthree) (1)

Florence2Run (1)

FluxGuidance (1)

GetNode (21)

Int-🔬 (1)

ModelSamplingFlux (1)

Note (1)

SetNode (17)

Custom Nodes (43)

ComfyUI

  • - ControlNetLoader (2)

  • - PreviewImage (4)

  • - BasicGuider (1)

  • - KSamplerSelect (1)

  • - VAEEncode (1)

  • - SamplerCustomAdvanced (1)

  • - RandomNoise (1)

  • - BasicScheduler (1)

  • - UpscaleModelLoader (1)

  • - VAELoader (2)

  • - CLIPTextEncode (1)

  • - EmptyLatentImage (1)

  • - KSampler (1)

  • - DualCLIPLoader (1)

  • - VAEDecode (2)

  • - UNETLoader (1)

  • - CheckpointLoaderSimple (1)

  • - ConditioningZeroOut (1)

  • - LoadImage (2)

  • - LoraLoader (1)

  • - SaveImage (1)

  • - easy cleanGpuUsed (1)

  • - ImpactSwitch (1)

  • - LineArtPreprocessor (1)

  • - TextInput_ (3)

Hakkun-ComfyUI-nodes

  • - Multi Text Merge (1)

  • - ShowText|pysssss (2)

  • - PlaySound|pysssss (1)

  • - GetImageSize (2)

  • - UltimateSDUpscale (1)

  • - JWImageResizeByLongerSide (2)

Checkpoints (1)

hunyuan_dit_1.2.safetensors

LoRAs (1)

flux_室内现代风格.safetensors