As we all know, although MJ is obviously much more beautiful than Stable Diffusion , the lower limit of generated content is higher , and the average aesthetic is online; but it is expensive and requires a network environment . If Stable Diffusion uses the original large model to directly output the picture, the quality is very average, and it needs to debug the model and parameters, which is difficult to learn and master. However, individual users want to fine-tune Stable Diffusion, or the learning cost of various parameters is also very high. But after proficiency, the upper limit is extremely high, and the cost is lower than that of MJ .
So for designers, is there a lower learning cost, more convenient usage scenarios, and cheaper usage costs? The latest new plugin based on Stable Diffusion can meet the above requirements.
PICPIK.AI
project address
https://www.picpik.ai/
Project features
- Multi-person collaboration : Only one account is required, and 15 people can use it across platforms at the same time. Sharing model and image resources makes it easier to use.
- Smart Face Swap : AI automatically fuses two pictures and replaces the face in one picture with the face in the other picture.
- High-definition enlargement : Use artificial intelligence deep convolutional neural network (CNN) to enlarge pictures intelligently and losslessly for free.
- Background removal : AI automatically recognizes the background area, and quickly obtains a PNG with a transparent background with one click.
- Image layering : Use deep learning image processing technology to quickly and intelligently layer images according to screen content.
- Vector conversion : AI automatically converts bitmaps to vector graphics quickly and easily, scaling to any resolution without becoming blurry.
- Native language expression : use GPT and vector database technology to intelligently analyze and understand your prompt words. From English to Chinese, making expressions more fluent, natural and efficient.
- Customized Services : Full-featured and easy-to-use training capabilities. While strictly protecting data security, model training can be completed in 10 minutes. Still feel troublesome? The model market also provides model trading and customization services! After testing, the face model training for a single person can be completed within 10 minutes.
AI access PS plug-in
Seamless access to workflow : The plug-in form is embedded in commonly used software such as Photoshop, deeply integrating the original capabilities of AI and software. Get started faster and seamlessly integrate into your workflow! Abundant reference image options : It realizes complete Controlnet and native reference image functions, and greatly simplifies the user experience. It can meet the needs of graph generation in various scenarios.
Deeply Optimized Preset Models
Deeply optimized preset models, the quality is higher! Constantly polish and optimize preset models. Generate highly expressive images easily without going through the learning curve and mastering various advanced skills!
online test
Here we choose the anime style and enter the prompt word: Ito Junji. Select a ratio of 1:1 and click Generate directly. The target picture can be generated in a short time, and the effect is still very good. After the generation is completed, you can download it directly, zoom in, replace the face, download the ps plug-in, etc.
Generate images using ps plugin
(1) Model selection
Before generating images, you first need to select an appropriate model based on the expected generated content. PICPIK provides 6 pre-installed models by default, which can cover most image generation scenarios. If you have special needs, you can train your own models through the training module, and these models will appear under the "Own Models" category. Tips: The landscape model can not only generate realistic scenery, but also can be used to generate sci-fi style pictures.
(2) Prompt words
Prompt words are the core modules needed to generate pictures. Traditional painting uses a brush to generate the expected picture, and the prompt word is like a brush in the field of AI image generation, requiring you to use language to express the generated picture.
How to express the content of the screen
There are two main ways to express, complete long sentences and keywords. You can use complete long sentences to express the content of the screen as in daily speech. When the screen is complex, or you don't have a particularly specific idea, you can also use comma-separated keywords to describe the screen. In fact, keywords may be the more mainstream method of use at present, because it is easier to adjust and control. Below are two examples of landscape screens in different usages, you can try them yourself. long sentence:
The aftermath of the setting sun rises in the sky in the distance, the vast sea is quiet and windless, hundreds of wind turbines stand row upon row on the mountains, the sky darkens little by little, countless sunsets are like fireworks, from The splendor fades gradually. Then the night descended on the bay without haste, and the harbor was lit up one after another.
Key words:
Sunset, calm sea, wind turbines on the hill, hundreds of thousands, sunset, bay, harbor, building lights, night
negative cue
Generally speaking, we only need to enter the content of the screen in the input box of the positive prompt words. But in some cases, in order to control the screen, we want to avoid the appearance of some specific content, at this time, we can use the negative prompt word input box. AI will try to avoid the content expressed by negative prompts from appearing on the screen.
multilingual switch
At present, most AI painting models are trained based on English prompts, so other languages cannot be used as prompts under normal circumstances. After turning on the multi-language switch, PICPIK will combine GPT and vector database technology to intelligently process and analyze your input, and can support mixed input of Chinese and various languages.
Dictionaries and auto-completion
Click the document icon in the upper right corner of the prompt word input box to open the prompt word dictionary. When you can't find a suitable expression, you can try to use the vocabulary in the dictionary to help express the content of the screen or improve the quality of the screen. In addition to actively opening, when entering prompt words, keywords in the dictionary will also appear below the input cursor as auto-completion content.
advanced grammar
Some specific grammars can be used to emphasize, weaken or even mix prompt words, and the related grammars are listed as follows: Note: The colons, parentheses and square brackets in the examples are all half-width English symbols. After enabling multi-language, if you use Chinese input, the system will automatically convert the corresponding Chinese full-width symbols. The concept of example effect a (pen) pen is strengthened to 1.1 times of the original. The concept of a ((pen)) pen is strengthened to 1.21 times of the original. Each additional group of brackets is 1.1 times more. The concept of a [pen] pen is weakened to the original 1/1.1. The concept of a [[pen]] pen is weakened by 1/1.21 times. Each additional group of brackets is 1.1 times more. The concept of a (pen: 2) pen is strengthened to 2 times of the original. The concept of a (pen: 0.5) pen is weakened to 50% of the original. The first 50% of the steps of a [pen: arrow: 0.5] A pen is generated, and arrows are generated in the last 50% of the steps. The result is a mixture of arrows and pens.
(3) Image generation operation
Generate images according to your settings. The bottom left of the generate button shows your remaining number of inferences, and the bottom right of the generate button shows the number of images generated at one time, which you can edit and modify. After clicking the "Generate Now" button, the progress of image generation will be displayed, and you can cancel it halfway, but the deducted number of inferences will not be restored. After completion, you can choose a picture to add to the canvas, and the selected picture can be switched repeatedly. Click "End" to return to the creation interface, and click "Regenerate" to regenerate the image using the current settings and prompt words.
(4) Generate graph
If you need to generate a picture similar to an existing image, you can use the existing image as a reference image. The reference picture can be obtained through external upload, or the content of the picture on the current canvas can be extracted by using the PS selection area. The reference strength in the reference picture module refers to the weight of the reference picture. If the reference strength is 100%, a picture that is exactly the same as the original picture will be generated. Using the "reverse description" function in the reference image module, the possible prompt words of the current picture will be automatically deduced and filled in the forward prompt word input box. When using a reference image, you can either leave the prompts blank or include some additional prompts. The supplementary prompt words will be used together with the reference picture as the guidance content for generating the picture.
Complete transparency
Complement the transparent pixel area. If there is a selection area, only the transparent area within the selection area will be completed. If there is no selection area, the transparent area in the entire canvas will be completed. The "reference intensity" in this function module is similar to the "reference intensity" in the "reference image" function, but the reference content is the picture around the transparent pixel.
（5） ControlNet
ControNet is a widely used mainstream reference map plug-in. Using its functions, it can efficiently complete tasks in some specific scenarios. For details, see the official description(opens in a new tab). The usage method of ContrlNet is similar to the "reference image" function in the picture, and a reference image is required (the acquisition method is the same as the "reference image" function), but the purpose is different. PICPIK presets 4 shortcut options for you, corresponding to 4 common uses. The specific description is as follows: Preset Use Simple Graffiti Based on the simple graffiti in the reference, generate detailed picture content. Character pose Based on the pose of a person in a reference frame, generate frame content in the same pose. The animation line draft automatically completes the details and colors according to the white line in the reference picture to generate a complete picture. Mainly used to generate animation content. Space line draft is similar to anime line draft, but it is better at straight line and three-dimensional space, and it is mainly used to generate architectural and interior design renderings. In addition, you can also refer to the official instructions to customize ControlNet parameters based on your usage scenarios and needs.
(6) Generate settings
Advanced parameters used when generating images. We have configured optimal parameter settings for the preloaded models. If you do not understand the relevant technical concepts, it is recommended to use the official settings directly without modification.
(7) Output settings
In Photoshop, the size of the output image is determined by the selection or canvas size. If the size of the output image exceeds the maximum limit set by the package, oversampling will be enabled by default, that is, the high-definition zoom function. You can also manually turn off this feature.
(8) Functional navigation
Functional navigation is mainly divided into 4 modules, creation, post-processing, history gallery and personal center. The authoring panel contains all the generating options. Post-processing includes various post-secondary editing functions. The history gallery shows all historical picture information and operations.
Product Pricing
The price is also very conscientious, and friends who need it can experience it.