[ComfyUI]HiDream-I1:ComfyUI原生支持!GGUF低显存可体验,文生图与图生图一览

img

HiDream-I1简介

在之前的文章已经多次介绍最新的17B文生图模型:HiDream-I1 。最新社区进展已经出现了一批基于HiDream-I1文生图 底膜的LORA模型,同时也得到了ComfyUI官方的原来支持。今天将在继续深入探索ComfyUI原生支持下的文生图和图生图工作流体验。

GitHub:https://github.com/HiDream-ai/HiDream-I1

HiDream-I1文生图ComfyUI体验

本文将使用ComfyUI本体工作流体验。文末网盘模型下载

• 更新ComfyUI本体即可,无需下载其他插件安装

• 模型下载地址:https://huggingface.co/Comfy-Org/HiDream-I1_ComfyUI/tree/main 、 https://huggingface.co/city96/HiDream-I1-Dev-gguf/tree/main

图片

图片

img

HiDream-I1文生图ComfyUI体验

HiDream-I1文生图ComfyUI体验工作流下载:
- RunningHUB文生图在线体验:https://www.runninghub.cn/post/1912548267165712385/?inviteCode=kol01-rh059
- RunningHUB图生图在线体验:https://www.runninghub.cn/post/1915248465113452546/?inviteCode=kol01-rh059

img

img

注意:

• 具有dev、full、fast3个版本,对应的采样步骤、CFG设置不一样。官方推荐设置:HiDream Full:hidream_i1_full_fp16.safetensors:shift: 3.0;steps: 50;sampler: uni_pc;scheduler: simple;cfg: 5.0。HiDream Dev:hidream_i1_dev_bf16.safetensors: shift: 6.0;steps: 28;sampler: lcm;scheduler: normal;cfg: 1.0 (no negative prompt)。HiDream Fast:hidream_i1_fast_bf16.safetensors:shift: 3.0;steps: 16;sampler: lcm;scheduler: normal;cfg: 1.0 (no negative prompt)

• 对于低显存也可以使用GGUF版本或NF4 fast版本,大约15G左右

• 更多图文和视频ComfyUI工作流参见个人主页:https://www.runninghub.cn/user-center/1890418187312222210?utm_source=kol01-RH059

01.文生图-玩偶展示

Create a toy in a blister of A rugged full figure ship skipper stands confidently in a medium shot,wearing weathered pale nautical gear—a faded cap, sturdy low boat shoes, and a Henry Lloyd red jacket. His gaze is steady, framed by deep wrinkles. Let it be an action figure. Next to the figure, there should be the toy's equipment, each in its individual blisters. 1) a slice of pizza. 2) lifesaver. 3) life jacket. Don't repeat the equipment under any circumstances. The card holding the blister should be blue navy. Also, on top of the box, write 'MAURO' in bold handwritten font. The figure and equipment must all be inside blisters. Visualize this in a realistic way.

最新HiDream-ComfyUI官方文生图

02.文生图-儿童画

Create a comical pencil drawn stick figure of a man on a blank sheet of white plain paper with a smile on its comical face. 

img

03. 图生图-街机游戏

1 chinese women, The image depicts an arcade environment, showcasing classic gaming machines with brightly lit displays. Prominent among the machines are two "Street Fighter" arcade cabinets, which indicate the retro gaming theme of the location. The subject in focus is a woman standing confidently near the arcade machines, dressed in an eye-catching outfit that includes a fitted white crop top, a black leather mini skirt, long black gloves, and knee-high black boots with red soles. Her sleek, long black hair cascades down, contributing to an edgy yet stylish appearance. Of note, she is positioned with one leg raised onto the arcade console, adding a dynamic, confident pose. The overall setting suggests a nostalgic yet modern vibe, combining retro gaming with contemporary style.In the background, several other arcade players can be seen seated, engaged with the machines, contributing to the lively atmosphere. The lighting of the arcade, complete with glowing screens and signage, enhances the vibrant ambiance of the space. The arrangement of the arcade machines, stools, and players immerses the viewer into the setting, hinting at a fondness for retro gaming culture.

img

04.图生图-金属人

The image depicts a menacing, otherworldly figure crafted entirely from sharp, metallic blades. Its humanoid form is engulfed in spikes and jagged edges, giving the impression of a living weapon. The imposing creature is adorned with massive wings, composed of countless thin, blade-like feathers, outstretched to emphasize its dominance and threat. Its entire body gleams under light, reflecting metallic hues ranging from silver to dark gray. The symmetrical arrangement of the spikes suggests an intentional design rather than chaotic randomness. Its glowing golden eyes pierce through the darkness, providing an eerie contrast to the cold, steel-like texture of its body. The eyes convey an ominous presence, hinting at intelligence or a supernatural essence. The backdrop is entirely black, emphasizing the figure’s layered sharpness and isolating it as the focal point of the composition.This figure resembles an abstract fusion of a dark angel and a biomechanical entity, evoking imagery similar to characters from dark fantasy or science fiction genres. Its design could align with creatures found in works such as *Final Fantasy*, *Warhammer 40,000*, or a villain from a futuristic dystopia. The scene conveys themes of danger, power, and cold, metallic beauty.

img

为了帮助大家更好地掌握 ComfyUI,我花了几个月的时间,撰写并录制了一套ComfyUI的基础教程,共六篇。这套教程详细介绍了选择ComfyUI的理由、其优缺点、下载安装方法、模型与插件的安装、工作流节点和底层逻辑详解、遮罩修改重绘/Inpenting模块以及SDXL工作流手把手搭建。

由于篇幅原因,本文精选几个章节,详细版点击下方卡片免费领取

一、ComfyUI配置指南

  • 报错指南
  • 环境配置
  • 脚本更新
  • 后记

img

二、ComfyUI基础入门

  • 软件安装篇
  • 插件安装篇

img

三、 ComfyUI工作流节点/底层逻辑详解

  • ComfyUI 基础概念理解
  • Stable diffusion 工作原理
  • 工作流底层逻辑
  • 必备插件补全

img

四、ComfyUI节点技巧进阶/多模型串联

  • 节点进阶详解
  • 提词技巧精通
  • 多模型节点串联

img

五、ComfyUI遮罩修改重绘/Inpenting模块详解

  • 图像分辨率
  • 姿势

img

六、ComfyUI超实用SDXL工作流手把手搭建

  • Refined模型
  • SDXL风格化提示词
  • SDXL工作流搭建

img

由于篇幅原因,本文精选几个章节,详细版点击下方卡片免费领取

img

在这里插入图片描述

### 文本像技术概述 文本像(Text-to-Image, T2I)是一种将自然语言描述转化为视觉表示的技术。为了实现这一目标,T2I方法通常依赖于深度学习模型,特别是成对抗网络(GANs)、变分自编码器(VAEs),以及其他先进的架构如扩散模型。 对于条件变量增强的T2I方法而言,通过引入额外的信息源可以显著提升成效果的质量和准确性[^1]。这些附加信息可能包括但限于场景、对话上下文、对象属性、边界框位置以及人体姿态的关键点等。每种类型的补充资料都有助于更精准地捕捉输入文本所传达的具体含义,从而指导模型创建更加贴合预期的画面。 #### 基于同形式辅助信息的方法实例 - **基于场景**:利用语义结构化表达来定义物体间的关系及其布局; - **基于对话**:考虑多轮交互中的连续性和一致性; - **基于属性驱动**:强调特定特征的重要性,比如颜色、形状或材质质感; - **基于边界框标注**:提供关于前景主体的位置提示; - **基于关键点**:用于描绘人物姿势或其他具有固定形态的对象的姿态变化。 上述各种方式均有助于提高最终产出物的真实感合理性,使得机器创作的作品仅限于表面相似度上的匹配,在深层次上也尽可能接近人类理解层面的要求。 除了注重如何有效融入外部线索外,确保成结果具备足够的精细程度同样是可忽视的一环。这意味着要对诸如纹理细节、光影效果乃至色彩平衡等因素加以严格把控,力求达到令人信服的艺术水准[^2]。 由于整个流程涉及海量的数据集训练及密集型运算任务的支持,故而离开强大算力资源作为保障基础。具体来说就是指那些配备了高端GPU集群的工作站或是云端服务平台所提供的分布式并行处理能力[^3]。 ```python import torch from diffusers import StableDiffusionPipeline model_id = "CompVis/stable-diffusion-v1-4" device = "cuda" pipe = StableDiffusionPipeline.from_pretrained(model_id).to(device) prompt = "A fantasy landscape with mountains and a lake under starry sky." image = pipe(prompt).images[0] image.save("fantasy_landscape.png") ``` 这段Python代码展示了使用预训练好的Stable Diffusion模型根据给定的文字提示`"A fantasy landscape with mountains and a lake under starry sky."`成一幅幻想风景画的过程,并保存为PNG文件格式。
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值