Playground is one of the best AI image generators. A very intuitive interface, many features to suit all needs and the best thing is that it’s free. The free account on Playground allows you to generate 500 images per day, but if you have larger needs and access to Premium features, you have subscriptions that start at $15 per month.
In this guide, I’ll cover the different aspects of the interface, which they do knowing that Playground AI uses Stable Diffusion. Therefore, knowing how to use Playground will allow you to understand how Stable Diffusion works in general and you will be able to apply it either on other platforms, or via a local installation if you have a good PC.
The Playground AI Board
There are two ways to generate an AI image on Playground AI, via the Board which is the classic interface or the Canvas which allows you to do Inpainting, Outpaing, and make much more advanced compositions.
On the left side of the Board of Playground AI, you have:
Filter
This is the name given to Custom models by Playground AI. For example, you have Deliberate, DreamShapper, PhotoReal, etc. Each model is adapted to a particular image style, photorealistic, pixelart, digital painting, manga, etc. Depending on which base model you choose, SDXL, Playground v2 or SD 1.5, filters will be available.
Today, SDXL has become the standard for image quality and that is what I recommend. But Playground v2, which is the Custom model developed by Playground AI, based on SDXL, is worth a try. In short, AI generation involves playing with models and settings and therefore, a lot of trial and error before obtaining the desired result.
Prompt
This is the main area where you put your prompt. Simply describe what you want to generate. It has to be in English. A prompt can be in several parts. For example, “a beautiful woman with her hair blowing in the wind” or “a beautiful woman with her hair blowing in the wind, forest in the background, night, diffused lighting”.
A longer prompt does not necessarily mean a good quality image. The filters or template you use will also affect the prompt. If you choose a filter that is optimized for manga, then your beautiful woman will be a drawing and not a photorealistic one.
Prompt for the image on the left: “a beautiful woman with her hair in the wind”
Prompt for the image on the right: “a beautiful woman with her hair in the wind, forest in the background, night, diffused lighting”
The Expand Prompt
The Expand Prompt is something new that Playground AI launched in November 2023. As its name suggests, it expands and improves your basic prompt. You can put just one word, if you activate the Expand Prompt, the AI will improve this word by adding other elements of context and parameters to give an exceptional image.
The Expand Prompt is limited to 30 per month in the free version, paid accounts can use it unlimitedly. Many platforms offer AI-enhanced prompts, but Playground’s blew me away with its performance.
Prompt for the image on the left: “a roasted chicken”
Prompt for the image on the right (with Expand Prompt enabled) “Roasted chicken resting on a rustic wooden cutting board, surrounded by rosemary sprigs and half-sliced lemons, kitchen counter in soft focus background, warm overhead lighting enhancing the golden-brown crispy skin texture, digital painting, ultra realistic, tantalizing still life”
Exclude from Image
This is the area for the negative prompt. Playground AI will automatically add negative words like ugly, deformed, missing hands, etc. Stable Diffusion, even if it has become very efficient, can still generate real monstrosities, the negative prompt makes it possible to reduce or avoid these artifacts.
This can also give more precision to the image you want to generate. For example, if you are looking for a photorealistic image, you can put 3D, cartoon, comics or manga in the prompts so that it focuses on photorealism.
Image to Image
Image to Image allows you to use a reference image to guide the AI in generating the image. Be careful, this is not in ControlNet where you have precise control over the image. Image to Image allows you to give a global reference. This option has a value which can go from 0 to 100. At 0, the reference image will not be taken into account at all, at 100, you will have an image identical to the base one. You have to play with the settings to obtain variations of an image that you already like.
Image To Image has a second function which is to improve an AI image that you like quite a bit, but you want to correct some defects. Because in Image to Image, the AI will again process your prompt using your base image and therefore, you have a multi-pass process that will smooth and beautify the image with each pass. The key is to find the right setting so that the image remains faithful to the original while providing variations that go well.
Left Image – Used in the Base Image with the prompt “a roasted lamb”
Right image – The result, we see roasted lamb, but which keeps the same style and structure as that of roasted chicken.
Likewise, you can use masks in Image to Image. When you put an image in Image to Image you will have an Edit with Mask option, click on it and the interface will change slightly. And you can draw a mask. The mask means that only the masked areas of the image will be affected by your prompt.
In Image to Image and mask mode, you should favor simple prompts of one or two words. Below, the image of the roasted chicken in Image to Image where I masked the lemons and replaced them with cherry tomatoes. Image to Image mode is very powerful and allows you to make small adjustments to an existing image.
Please note that you do not need to use an AI image for image to image, you can use any image and use this mode with a mask or not. A small downside is when you switch to Image to Image and mask mode, Playground AI switches to the SD 1.5 model which is less efficient than SDXL, but for small changes, it is sufficient.
Control Traits
Control Traits is Playground AI’s implementation of ControlNet. ControlNet is one of the biggest revolutions in AI imaging in 2023 because it allows you to give control to the image. Image AI can generate very beautiful images, but we had no precision on the result. ControlNet partially solves this problem, by making it possible to analyze the different characteristics of an original image and apply these characteristics to the AI image.
At the beginning, it was mainly the position and that’s what Playground AI offers. But today, you can take the style of an image and apply it to the result. Playground offers Edge, Depth and Pose. Edge which is known as Canny in real ControlNet will analyze the edges of an image and apply it to your AI image. Depth attempts to differentiate between the background and foreground of the image. And Pose is a free interpretation of the original image.
If you are looking for a pose that is as precise as possible, then use Edge, if your image has perfect contrast between the background and the foreground, then choose Depth. And if you like the original pose, but want the AI to interpret it its way, then choose Pose. Note that I had problems displaying the Control Traits window in Chrome.
But it works on Firefox. Below you have an original image which is an angel on one hand and its three interpretations by Edge, Depth and Pose from Control Traits. Knowing that when you activate Control Traits, you have settings to modify the intensity with which it will apply the control. In these three images below, the intensity is at 0.5.
On the right side of the Board of Playground AI, you have:
Model
These are the different versions of Stable Diffusion that are available on Playground AI. You have SDXL which is the latest version, Playground v2 which is their own model and SD 1.5 which is the old version before SDXL.
The standard today is to use SDXL and if the result is not up to par, then we can combine it with the filters we saw above. One of the particularities of SDXL is that you can have a Refined version. You have a specific setting for refining a little lower.
Image Dimensions
Very simple adjustment, it allows you to select the size of the generated image. You have square format, 4:3, 3:4, etc. Knowing that it’s just the base generation. But once the image is generated, you can Upscale it to resize it by 4. The Upscale is limited to 30 per day for free accounts.
Prompt Guidance
Known as the CFG in Stable Diffusion, Prompt Guidance in Playground AI determines how strongly the AI will respect your prompt. If the value is very high, then the AI will favor promptness over creativity. The golden rule with Prompt Guidance is to use a fairly basic value and let the AI be creative.
The norm was to use a value of 7, but now, with models like SDXL Turbo (which speeds up image generation enormously), you can even use a Prompt Guidance of 3 and it will work.
Quality & Details
This setting is equivalent to Steps in software like Automatic1111. It determines the number of passes the AI will work on before generating the final image. Obviously, the higher the value, the more detail there is, but obviously it depends on more time. A value of 50 is more than enough on Playground AI.
Refinement
This is the setting we talked about a little above on SDXL. It is only available for this version of Stable Diffusion. Refinement will add detail to the image and it always adds a little. However, just because you have a Refinement of 100 does not mean that it will magically generate an exceptional image.
And the proof below or from left to right, you have a Refinement of 0, 17 and 100. The last image is totally distorted. The rule is to use a low level of Refinement and move up gradually.
Prompt : “a luxury watch” with refinement of 0, 17 and 100 (from left to right)
Seed
The Seed is a numerical value assigned to each image generated by Stable Diffusion. You must always activate Random Seed in Playground AI to have a different image each time. If you use the same Seed, then obviously you will have the same image. This could be useful at a certain time since we used hacks with masks to have consistent faces from one AI generation to the next.
Sampler
The Sampler is a setting that you don’t need to touch. It is related to how Stable Diffusion generates the final image from noise. At first there is just noise in the image and then at each step SD will “denoise” this image to provide the result you want.
It samples a new image at each step and that’s why we say it does Sampling. You have lots of different Samplers such as Euler A, DDIM, Heun, Karras, etc. Some users will prefer one Sampler over another.
I have always used Euler a with good results. Playground AI offers many Samplers and some are reserved for paid users. But the free ones are more than enough. And the Sampler is not the only factor that determines the final image. So don’t worry about it.
Number of images
Very simple, it’s the number of images you generate each time. The maximum is 4 for the free version, 8 for the Pro versions and they even have a Turbo version that allows you to generate 16 images at the same time. 4 images is more than enough. Knowing that basic AI image generation is just the first step. You have Image to Image to improve it later.
Private Session
On Playground AI, all your images are public by default. This way, the community will be able to see them and vote for them. Small aside, don’t try to generate NSFW images, violent images or that kind of thing, because Playground AI has very powerful filters that detect the slightest misbehavior.
Initially, the image will not be generated, but if they detect abuse of their service by generating prohibited images, then they may suspend your account. If you have a subscription, you can enable this option so that your images are private by default. No need to reactivate it every time.
Options after you generate your AI images
Once you have generated your image, you still have plenty of options available:
- Edit: Edit mode is what we saw with Image To Image. You can request small modifications by adding masks if necessary.
- Copy Link: Copies the image link to display it on Playground AI. It’s the equivalent of going to your Profile to access all the images you have generated
- Edit in Canvas: Edit the image in Canvas. We cover Canvas a little further down.
- Make Private: Make the image private.
- Download: Download the image
- Inpaint: Equivalent to using a mask with Image to Image mode.
- Use Image to Image: Send the image in Image To Image.
- Create Variants: The same as Image to Image, but this will directly generate new AI images using this one as a base point.
- Face Restoration: Allows you to correct facial defects. With SDXL now generating perfect faces, this option is unnecessary. It was mainly used at the time of SD 1.5.
- Upscale By 4x: Resizes the image by 4. You are entitled to 30 Upscale per day in the free version.
- Remove Background: Removes the background from the image.
- View Full Screen: View the image in full screen.
- Delete: Delete the image. This is useful if your images are still public and you have created some that are particularly bad.
The Playground AI Cancas
As we can see, the classic generation of an AI image via the Board gives a lot of possibilities with Playground AI. But the latter also offers Canvas where you can make complex compositions, do Outpainting, Inpainting, dynamically delete elements from the image and lots of other things. Think of it as AI-powered photo editing software. In the free version, you are entitled to 3 Canvas projects.
We imported the image of our little boy into Canvas. The main menus are at the top and you have:
- Import Image: To import an image into the Canvas. It is not necessary that you generated it with Playground AI. You can edit any image.
- Move and Select: To move and select images.
- Hand Tool/Pan: To move around the Canvas.
- Generate Image: The heart of the Canvas where you generate the AI image. Knowing that the interface will change depending on the mode you select. In Pan and Select mode, the interface will be much cleaner while the AI generation, we have a classic interface.
- Draw: To draw directly in the image. This option has two modes, Simple Pencil and Draw to Edit. The first allows you to draw in classic mode. Draw to Edit allows you to draw and you have a prompt interface to generate the result according to your drawing.
- Erase: Allows you to erase part of the image. It also has two modes, Simple Eraser and Object Eraser. With the latter, you can mask a specific part of the image to erase it. The result is variable and depends on the object, background, etc.
- Inpaint: Allows you to hide part of the image and add modifications or other. It also has two modes, Inpaint and Instruct to Edit. The latter is similar to what we saw in the Board. You hide part of the image and you can give simple instructions.
- Comment: You can put comment points on the image. This is especially useful if you use the Canvas collaboratively, because yes, you can work with several people in the Playground AI Canvas.
Outpainting and Inpainting in Playground AI
One of the great strengths of Canvas compared to Board is that you can do Inpainting and Outpainting. The first was also possible in the Board. But Outpainting is only possible in the canvas, because it allows you to extend an existing image. It is mainly a web-ui like Invoke AI that popularized the concept.
To use it, you switch to Generate Image mode and make the frame fill part of the image, but not all of it. You can use Shift key + mouse to keep the proportions. Likewise, when your frame aligns perfectly with the image, it will turn red to indicate perfect alignment.
In Outpainting, the essential thing to understand is that the AI simply “sees” what is in the frame. He doesn’t see half of the boy, because he’s out of the frame. But overlapping it in this way allows the AI to understand part of the original image and extend it. You don’t need a prompt for Outpainting.
With the prompt empty, click Generate to expand the image. Note that rather than clicking the Generate button each time, you can also use the Ctrl + Enter shortcut key combination to start the generation. And this is valid for the Canvas and also the Board.
Above is our image which has been extended on both sides using Outpainting. Outpainting gives almost infinite possibilities in AI image generation and especially with SDXL and current versions, we have almost no demarcation which shows that we are composing an image. In the past, you often had lines appearing. This allows you to create large images in length and width starting from an original image. Obviously, avoid making images that are too large, as this will weigh down the interface with the risk of bugs.
If the frame is in the image, then the Canvas will automatically switch to Inpainting mode and it works as explained with classic Inpainting. You can add masks, make small modifications, etc. I personally find that Outpainting is much more efficient than Inpainting on the canvas.
Once you’re done, you go to Selection mode and there you have other options going from left to right to upload image, remove background, make image to image, crop and delete the image. Next to the Delete option, you have three dots which give you other actions such as bringing the image to the front or foreground.
Because yes, the Playground AI Canvas can have types of layers. You can flip the image vertically or horizontally, use it as Control Traits, Upscale, etc.
The big final word about Playground AI
As we saw in this enormous file, behind its simple tool appearance, Playground AI is a real image studio, all assisted by AI. The interface is very well thought out whether for the Board or the Canvas. The options are dynamic and fluid and adapt to what you want to do to the image.
The amount of options is impressive and the best part is that much of it is available for free. If you have money, don’t hesitate to take out a paid subscription, because these days, there are plenty of AI tools that suck money out of you for basic options.
Playground AI does not make fun of you and it offers you a complete free version and the paid versions offer you total freedom whether in the creation of AI images, but also the functionalities available whether in terms of dimension, d improvement and composition. When I say that Playground AI is the best AI image generator on the web, my daily use proves it.
The amount of options can make you dizzy. I advise you to familiarize yourself with the Board and even with little knowledge, you will be able to generate your first AI images quite easily. In addition, they also have a Discord where they regularly announce new products without forgetting that you will have a community active enough to help you in the event of a problem.
0 Commentaires