如何在功能较弱的PC上运行强大的Flux AI图像生成器

币界网Published on 2024-08-12Last updated on 2024-08-13

币界网报道:

当Flux几天前突然出现时,它很快就赢得了开源图像生成器皇冠上的明珠的声誉。它与Midjourney的美学能力相匹配,同时在快速理解和文本生成方面绝对压倒了它。捕获?你需要一个拥有超过24GB VRAM(甚至更多)的强大GPU才能让它运行。这比大多数游戏设备都要强大,更不用说你的普通工作笔记本电脑了。

但是,人工智能社区从不放弃挑战,他们卷起集体袖子开始工作。通过量化的魔力——一个压缩模型数据的花哨术语——他们设法将Flux缩小到更易于管理的大小,而不会牺牲太多的艺术魔力。

让我们把它分解一下:最初的Flux型号使用了全32位精度(FP32),这就像驾驶一级方程式赛车去杂货店——对大多数人来说都是多余的。第一轮优化为我们带来了FP16和FP8版本,每个版本都以一点点的准确性换取了效率的大幅提高。FP8版本已经改变了游戏规则,让拥有6GB GPU(想想RTX 2060)的人加入了这场盛宴。

为此,您需要禁用系统内存回调以实现稳定扩散,这样您的GPU就可以将其部分工作从内部VRAM卸载到系统RAM。这避免了臭名昭著的OOM(内存不足)错误,尽管其运行速度要慢得多。要禁用此选项,请按照Nvidia的教程进行操作。

但要戴上帽子,因为它会变得更好。

人工智能世界的真实MVP进一步突破了界限,发布了4位量化模型。这些坏男孩使用一种叫做“正常点”(NP)量化的东西,它提供了一个质量和速度的最佳点,让你的土豆电脑感觉像是得到了涡轮增压。NP量化不会像FP量化那样降低质量,因此一般来说,运行此模型可以在高速下获得很好的结果,只需要很少的资源。

这几乎好得令人难以置信,但这是真的。

在fp8和np4中使用Flux Dec获得的结果。同样的提示,不同的种子。

如何在低端GPU上运行Flux

那么,你实际上是如何运行这个简化版本的Flux的呢?首先,你需要一个像SwarmUI、ComfyUI或Forge这样的界面。我们喜欢ComfyUI的多功能性,但在我们的测试中,Forge的速度比其他界面提高了约10-20%,所以这就是我们在这里使用的。

前往Forge GitHub仓库(https://github.com/lllyasviel/stable-diffusion-webui-forge)并下载一键安装包。它是开源的,经过社区审查,所以这里没有粗略的业务。

对于NP4 Flux型号本身,Civit AI是您的一站式服务。你有两种口味可供选择:Schnell(速度)和Dex(质量)。两者都可以从这个页面下载。

一旦你下载了所有内容,就到了安装时间:

    解压缩Forge文件并打开Forge文件夹。运行update.bat以获取所有依赖项。启动run.bat以完成设置。

现在,将这些闪亮的新Flux模型放入Forge安装中的webuimodelsStable扩散文件夹中。刷新Forge web界面(如果你感觉老派,可以重新启动),然后砰的一声——你在做生意。

专业提示:要真正从复活的装备中挤出每一滴性能,请降低分辨率。与其选择全SDXL(1024x1024)分辨率,不如尝试更适中的SD1.5尺寸(768x768、512x768等)。你总是可以稍后升级,并使用Adetailer来处理那些脆脆的细节。

让我们谈谈数字:在配备6GB VRAM的RTX 2060上,Flux Schnell在NP4模式下可以在大约30秒内生成512x768的图像,而FP8版本需要107秒。想做大吗?用高分辨率修复将那个坏男孩升级到1536x1024大约需要五分钟。

想在不破坏GPU的情况下做大吗?更好的选择是从SD1.5分辨率的Flux Schnell开始,然后通过img2img发送该创建。使用低降噪强度的标准稳定扩散模型(SD1.5或SDXL)进行升级。整个过程大约需要50秒,与MidJourney在缓慢的一天中的输出相当。您将在不熔化图形卡的情况下获得令人印象深刻的大规模结果。

真正的踢球者?据报道,一些疯狂的家伙让Flux Schnell NP4在GTX 1060上运行,VRAM为3GB,Flux Dev每次迭代需要7.90秒。我们谈论的是一个几乎依赖生命支持的GPU,它正在这里产生尖端的人工智能艺术。对于几乎符合养老金条件的硬件来说,这并不算太破旧。

Trending Cryptos

Related Reads

After Missing the 20x, I've Found a 'Dumb' Method for AI Investing

**Missing the 20x Opportunity: A Simple 'Dumb' Approach to AI Investing** The AI boom, driving NVIDIA's revenue from $60B to $216B in two years, creates immense investment pressure. However, like the internet bubble of 2000, the largest AI opportunities likely lie ahead, perhaps after a correction. Instead of rushing in now or waiting paralyzed for a crash, the author proposes a third way: building a "knowledge warehouse" by systematically mapping the AI industry to be ready when opportunities arise. The core of the strategy is understanding AI's four-layer value chain: 1. **Compute Infrastructure (The "Engine"):** This foundational layer, where all money eventually flows, includes: a) **Chip Design:** NVIDIA's dominance via its CUDA ecosystem, b) **Chip Manufacturing/Packaging/Memory:** TSMC's near-monopoly in advanced manufacturing and SK Hynix's lead in High Bandwidth Memory (HBM), c) **Optical Interconnects:** Essential for large-scale AI clusters (e.g., Lumentum, Coherent), d) **Cooling & Power:** Critical for high-density AI data centers (e.g., Vertiv), e) **Servers/Data Centers & Cloud Platforms:** The physical and virtual wholesale providers. 2. **Models & Tools (The "OS"):** The competitive layer of foundation models (OpenAI, Anthropic, Google, Meta, xAI), now generating real revenue. A key shift is the center of gravity moving from **Training** models to **Inference** (running models), which demands different chip characteristics and could challenge NVIDIA's monopoly. 3. **Middleware & Platform ("The Glue"):** Connects models and applications (e.g., Scale AI, Hugging Face). This layer could explode if applications take off. 4. **Vertical Applications ("The Cash Register"):** Where AI meets end-users (e.g., enterprise AI, coding tools, medical AI, robotics). A critical cross-cutting constraint is **Energy**, as AI's massive power consumption drives investment in nuclear and other energy infrastructure. The author identifies four key questions for further research: 1) How will the shift from Training to Inference reshape the competitive landscape? 2) With tech giants spending over $600B on capex, where is the ROI from AI applications? 3) What are the under-the-radar opportunities in the "second" and "third" circles of the value chain (e.g., cooling, specialty foundries)? 4) How will geopolitics (e.g., U.S.-China chip restrictions) bifurcate the supply chain? The conclusion is that missed opportunities stem from insufficient research, not slow timing. By methodically studying each layer—its business models, competition, and valuations—investors can build the "killer intuition" needed to act decisively when the market presents its chance.

marsbit16m ago

After Missing the 20x, I've Found a 'Dumb' Method for AI Investing

marsbit16m ago

Rented Faith: How Much of the Bitcoin ETF Inflows Is Real Money?

"Rented Conviction: How Much of Bitcoin ETF Flows Is Real Money" The weekly inflows into Bitcoin ETFs are often interpreted as a gauge of institutional belief. However, a significant portion of this activity is driven by a hidden arbitrage trade, not directional conviction. The core mechanism is a cash-and-carry arbitrage: traders buy spot Bitcoin (often via ETFs) while simultaneously shorting CME futures to lock in the price difference, or "basis." This delta-neutral trade is essentially an interest rate play. In weekly data, about half the fluctuation in ETF flows can be explained by new short positions added by leveraged funds (hedge funds), with a correlation of 0.70. Bitcoin's price movement in a given week shows no statistical power in predicting these flows. While this arbitrage trade drives weekly *volatility*, it is not the main component of the cumulative *stock*. Of the total ~$55 billion in net ETF inflows, the current net arbitrage position is only about $1 billion. The remainder is steady, directional buying averaging ~$400 million per week, which constitutes the vast majority of the accumulated "mountain" over two years. Thus, ETF flow data overstates the *volatility* of conviction, not its *level*. This arbitrage trade has been unwinding for nearly two years. Leveraged fund short positions peaked at ~$14 billion in late 2024 and have since declined to ~$4.5 billion. When the basis compresses to unprofitable levels, ETF inflows and short positions retreat together. Recent outflows should not be mistaken for a loss of faith but rather the routine unwinding of this rate trade. For Ethereum ETFs, the pattern is weaker. Accounting for staking yield makes the basis often negative, so neither strong conviction buying nor robust arbitrage supports its flows. To interpret ETF flows correctly, monitor the CME basis versus T-bill rates and leveraged fund net shorts. They reveal how much of the next "demand" headline is real. The real, patient buy-and-hold demand is what constitutes the enduring bulk of ETF assets.

marsbit1h ago

Rented Faith: How Much of the Bitcoin ETF Inflows Is Real Money?

marsbit1h ago

Trading

Spot
Futures

Hot Articles

How to Buy FLUX

Welcome to HTX.com! We've made purchasing Flux (FLUX) simple and convenient. Follow our step-by-step guide to embark on your crypto journey.Step 1: Create Your HTX AccountUse your email or phone number to sign up for a free account on HTX. Experience a hassle-free registration journey and unlock all features.Get My AccountStep 2: Go to Buy Crypto and Choose Your Payment MethodCredit/Debit Card: Use your Visa or Mastercard to buy Flux (FLUX) instantly.Balance: Use funds from your HTX account balance to trade seamlessly.Third Parties: We've added popular payment methods such as Google Pay and Apple Pay to enhance convenience.P2P: Trade directly with other users on HTX.Over-the-Counter (OTC): We offer tailor-made services and competitive exchange rates for traders.Step 3: Store Your Flux (FLUX)After purchasing your Flux (FLUX), store it in your HTX account. Alternatively, you can send it elsewhere via blockchain transfer or use it to trade other cryptocurrencies.Step 4: Trade Flux (FLUX)Easily trade Flux (FLUX) on HTX's spot market. Simply access your account, select your trading pair, execute your trades, and monitor in real-time. We offer a user-friendly experience for both beginners and seasoned traders.

1.7k Total ViewsPublished 2024.05.03Updated 2026.06.02

How to Buy FLUX

Discussions

Welcome to the HTX Community. Here, you can stay informed about the latest platform developments and gain access to professional market insights. Users' opinions on the price of FLUX (FLUX) are presented below.

活动图片