If you check out Intel’s AI Playground, which includes every thing from AI artwork to an LLM chatbot to even text-to-video in a single app, you may suppose: Wow! OK! An all-in-one native AI app that does every thing is value making an attempt out! And it’s… besides that it’s made for only a small slice of Intel’s personal merchandise.
Quite merely, no single AI app has emerged because the “Amazon” of AI, doing every thing you’d need in a single service or website. You can use a device like Adobe Photoshop or Firefly to carry out refined picture generations and modifying, however chatting is out. ChatGPT or Google Gemini can converse with you, even producing pictures, however to a restricted extent.
Most of those providers require you to hopscotch backwards and forwards between websites, nevertheless, and might value cash for a subscription. Intel’s AI Playground merges all of those inside a single, well-organized app that runs domestically (and fully privately) in your PC and it’s all free of charge.
Should I allow you to in on the catch? I suppose I’ve to. AI Playground is a showcase for Intel’s Core Ultra processors, together with its CPUs and GPUs–the Core Ultra 100 (Meteor Lake) and Core Ultra 200 (Lunar Lake) chips, particularly. But it might be so, so a lot better if everybody may use it.
Mark Hachman / Foundry
Yes, I understand that some customers are fairly suspicious of AI. (There are even AI-generated information tales!) Others, nevertheless, have discovered that sure duties of their each day life corresponding to business email can be handed off to ChatGPT. AI is a device, even when it may be utilized in methods we disagree with.
What’s in AI Playground?
AI Playground has three essential areas, all designated by tabs on the highest of the display:
- Create: An AI picture generator, which operates in both a default text-to-image mode, or in a “workflow” mode that makes use of a extra refined again finish for higher-quality pictures
- Enhance: Here, you may edit your pictures, both upscaling them or altering them by way of generative AI
- Answer: A standard AI chatbot, both as a standalone or with the flexibility to add your personal textual content paperwork
Each of these sections is what you may name self-sufficient, usable by itself. But within the higher right-hand nook is a settings or “gear” icon, which comprises a terrific variety of extra choices, that are completely value inspecting.
How to arrange and set up AI Playground
AI Playground’s power is in its thoughtfulness, ease of use, and ease. If you’ve ever used an area AI software, you realize that it may be tough. Some features are content material with only a command-line interface, which can require you to have a working information of Python or GitHub. AI Playground was designed across the premise that it’ll handle every thing with only a single click on. Documentation and explanations is perhaps a bit of missing in locations, however AI Playground’s ease of use is unparalleled.
AI Playground might be downloaded from Intel’s AI Playground page. At press time, AI Playground was on model 2.2.1 beta.
Mark Hachman / Foundry
Note that the app and its back-end code require assist for both a Core Ultra H (a “Meteor Lake” chip, the Core Ultra 200V) or both of the Intel Arc discrete GPUs, together with the Alchemist and Battlemage elements. If you personal a large gaming laptop computer with a 14th-gen Intel Core chip or an Nvidia RTX 5090 GPU, you’re out of luck. Same with the Core Ultra 200H or “Arrow Lake.”
Since that is an “AI Playground,” you may suppose that the chip’s NPU could be used? Nope. All of those purposes faucet simply the chip’s built-in GPU and I didn’t see the NPU being accessed as soon as through Windows Task Manager.
Also, understand that the GPU’s UMA body buffer, the reminiscence pool that’s shared between system reminiscence and the built-in GPU, is what these AI fashions depend upon. Intel’s built-in graphics shares half the out there system reminiscence with the system reminiscence, as a unified reminiscence structure or UMA. Discrete GPUs have their very own devoted VRAM reminiscence to tug from. The backside line? You might not have sufficient video reminiscence out there to run each mannequin.
Downloading the preliminary AI Playground software took about 680 megabytes on my machine. But that’s solely the shell software. The fashions require a further obtain, which is able to both be dealt with by the AI Installer software itself or might require you to click on the “download” button itself.
The good factor is that you just don’t must handle any of this. If AI Playground wants a mannequin, it can let you know which one it requires and the way a lot area in your exhausting drive it requires. None of the fashions I noticed used greater than 12GB of cupboard space and plenty of a lot much less. But if you wish to check out a variety of fashions, be ready to obtain a pair dozen gigabytes or extra.
Playing with AI Playground
I’ve known as Fooocus the easiest way to generate AI art on your PC. For its time, it was! And it really works with nearly any GPU, too. But AI Playground could also be even simpler. The tab opens with simply the area for a immediate and nothing else.
Like most AI artwork, the immediate defines the picture and you may get actually detailed. Here’s an instance: “Award winning photo of a high speed purple sports car, hyper-realism, racing fast over wet track at night. The license plate number is “B580″, motion blur, expansive glowing cityscape, neon lights…”
Mark Hachman / Foundry
Enter a immediate and AI Playground will draw 4 small pictures, which seem in a vertical column to the left. Each picture progresses in a collection of steps with 20 because the default. After the picture is accomplished, some small icons will seem subsequent to it with extra choices, together with importing it into the “Enhance” tab.
The Settings gear is the place you may start tweaking your output. You can choose from both “Standard” or “HD” decision, which adjusts the “Image Size” discipline. You can alter picture dimension and determination, and tweak the format. The “HD” possibility requires you to obtain a special mannequin, as does the ‘Workflow” option to the upper right, which adds workflows based on ComfyUI. Essentially, they’re simply higher wanting pictures with the choice to information the output with a reference picture or different workflow.
Mark Hachman / Foundry
For now, the default mannequin might be adjusted through the “Manual” tab, which opens up two extra choices. You’ll see a “negative prompt,” which excludes issues that you just put in, and a “Safe Check” to show off gore and different disturbing pictures. By default, “NSFW” (Not Safe for Work) is added to the unfavourable immediate.
Both the Safe Check and NSFW unfavourable immediate solely seem as choices within the Default picture generator and appear to be on by default elsewhere. It’s as much as you whether or not or to not take away them. The Default mannequin (Lykon/dreamshaper-8) has apparently been educated on nudity and celebrities, although I caught to public figures for testing functions.
Note that all your AI-generated artwork stays native to your PC, although Intel (clearly) warns you to not use an individual’s likeness with out their permission.
There’s additionally a jaw-droppingly apparent bug that I can’t consider Intel didn’t catch. Creating an HD picture typically begins its pictures with “UPLOAD” projected over the picture, and generally renders the ultimate picture with it on, too. Why? Because there’s a discipline so as to add a reference picture and UPLOAD is true in the course of it. Somehow, AI Playground used the UPLOAD font as a part of the picture.
Mark Hachman / Foundry
Though my check machine was a Core Ultra 258V (Lunar Lake) with 32GB of RAM, an 896×576 picture took 29 seconds to generate, with 25 rendering steps on the Default Mode. Using the Workflow (Line2-Image-HD-Quality) mannequin at 1280×832 decision and 20 steps, one picture took two minutes 12 seconds to render. There’s additionally a Fast mode which ought to decrease the rendering time, although I didn’t actually just like the output high quality.
If you discover a picture you want, you should utilize the Enhance tab to upscale it. (Upscaling is being added to the Windows Photos app, which is able to ultimately be made out there to Copilot+ PCs utilizing Intel Core Ultra 200 chips, too.) You may also use “inpainting,” which lets you re-generate a portion of the display, and “outpainting,” the approach which was used to “expand” the boundaries of the Mona Lisa portray, for instance. You may also ask AI to tweak the picture itself, although I had issues making an attempt to generate a passable outcome.
Mark Hachman / Foundry
The “Workflow” tab additionally hides some attention-grabbing utilities corresponding to a “face swap” app and a solution to “colorize” black-and-white pictures. I used to be disillusioned to see {that a} “text to video” mannequin didn’t work, presumably as a result of my PC was operating on built-in graphics.
The “Answer” or chatbot portion of the AI Playground appears to be the weakest possibility. The default mannequin, by Microsoft (Phi-3-mini-4K-instruct) refused to reply the dumb comic-book-nerd query, “Who would win in a fight, Wonder Woman or Iron Man?”
Mark Hachman / Foundry
It continued.
“What is the best car for an old man? Sorry, I can’t help with that.”
“What’s better, celery or potatoes? I’m sorry, I can’t assist with that. As an AI, I don’t have personal preferences.”
And so on. Switching to a special mannequin which used the OpenVINO programming language, although, helped. There, the OpenVINO/Phi-3.5-mini-instruct-int4 mannequin took 1.21 seconds to generate a response token, producing tokens to the tune of about 20 tokens per second. (A token isn’t fairly the size of a phrase, but it surely’s a superb rule of thumb.) I used to be additionally in a position to do some “vibe coding” — producing code through AI with out the faintest clue what you’re doing. By default, the output is just some hundred tokens, however that may be adjusted through a slider.
You may also simply import your personal mannequin, too, by dropping a GGUF file (the file format for inference engines) into the suitable folder.
Adapt AI Playground to AMD and Nvidia, please!
For all that, I actually like AI Playground. Some individuals are notably (justifiably?) skeptical of AI, particularly how AI could make errors and exchange the genuine output of human artists. I’m not right here to argue both aspect.
What Intel has accomplished, nevertheless, is create a surprisingly good general-purpose and fanatic software for exploring AI, that receives frequent updates and appears to be persistently bettering.
The smartest thing about AI Playground? It’s open supply, that means that somebody may in all probability give you a fork that permits for extra GPUs and CPUs to be carried out. From what I can see, it simply hasn’t occurred but. If it did, it might be the one unified native AI app I’ve been ready for.