Superb full-blooded no-charge AI drawing tutorials are here (online Stable Diffusion in one click)
I. Introduction
1.1 AI drawing
AI plotting is the process of entering some descriptive statements (Vincennes diagram) or pictures (Tucson (city)), AI can generate creative drawings based on input information.AI drawing is one of the hottest application areas of AI today, and is also a type of generative artificial intelligence AIGC.
Common AI drawing software is as follows.
(1) Literary Heart of One
Wenxin Yige is an AI art and creative assistance platform based on the Baidu Flying Paddle and Wenxin Big Model of the Wenborn Picture System. Through Wenxin Yige intelligent production of diverse creative pictures, and draw creative inspiration from them to break the creative bottleneck.
(2)Midjourney
Midjourney, which generates images based on text, enters public beta on July 12, 2022, simply by typing in a keyword and generating the corresponding image through an AI algorithm. Midjourney is populated with the artistic styles of numerous painters, such as Leonardo da Vinci, Dalí, and Picasso, and also recognizes a specific lens or photographic terminology.
Midjourney has a wide range of application scenarios and can be used in poster, IP design, advertising design, creative design, art education, comic creation and other fields.
(3)Adobe Firefly
Firefly, the creative generative AI introduced by Adobe on March 22, 2023, uses text to generate images, audio, illustrations, video, and 3D imagery.Firefly will provide new ways to ideate, create, and communicate while dramatically improving creative workflows.
(4)Stable Diffusion
Stable Diffusion is a text-to-image generation model based on Latent Diffusion Models (LDMs), which can generate high-quality, high-resolution, high-realistic images from arbitrary text input. Compared with traditional generation models, Stable Diffusion generates images with higher quality, faster speed and lower cost.
1.2 Stable Diffusion
This article is mainly shared based on Stable Diffusion because ↓
1.2.1 Brief description of the principle
As you can see from the name Stable Diffusion, this one mainly uses theDiffusion Model. In short, the diffusion model is the successive application of the denoising white encoder to generate images step by step.
The meaning of Diffusion is essentially aiterative process. While diffusion is the repeated addition of small, random noise to an image, diffusion modeling is the opposite of this process, generating high-definition images from the noise, and the neural network trained is usually a U-net.
1.2.2 Application process
The process is divided into 4 main steps:
- Selecting the rightmould
- Enter the exactclue
- parameterization
- Post-processing technology
(1) On the model
Stable Diffusion models are mainly divided into large models and fine-tuned models. The common large models are Checkpoint (ckpt) and safetensors, which are usually trained by Dreambooth, characterized by good graphical results, but the training process is slower and the generated model file is larger; while the fine-tuning model generally needs to be used in conjunction with the large models, which commonly includeLoramodel, VAE model, Embedding model and Hypernetwork model.
This article is shared primarily based onLora modelLora directly translates to low-order adaptation of a large language model. lora is a large file with a large amount of saved information, and it is more effective for character reduction, action specification, and painting style specification.
(2) About the cue word
Stable Diffusion cue words are categorized asPositive Cuesandinverse cue word, generally need to include [picture quality] + [picture style] + [picture subject] + [picture scene] + [other elements].
Second, AI drawing tools
2.1 Toast TusiArt
Tusi.Art is an AI image generation platform based on Stable Diffusion, with 100 arithmetic points per day, it can generate 100 images and can be downloaded in HD.
2.2 Mileage LibLibAI
LiblibAI basically 1:1 restore Stable Diffusion running interface (with some castration), running the most at a glance, simple and efficient interface, and very easy to get started. In addition, LiblibAI can run 100 AI charts for free every day, and the subsequent demonstration is mainly based on LiblibAI.
2.3 Native deployment
Local or cloud deployment of native Stable Diffusion, e.g. AliCloud cloud deployment↓.
III. One-click ready to use
3.1 Unpacking and Tasting
After registering and logging in to LibLibAI, click on [Online Graphics].
After entering the [Online Generate Picture] page, click [Generate Picture] directly to open the box and taste it.
3.2 Model correlation
Select a reference model on the home page and tap on it.
Click [Add to Model Library], then mouse over the [Generation Information] of the model, and copy the forward cue, reverse cue, and sampling method to the [Online Graph Generation] page.
In [Generate Picture Online], click [Generate Picture] after you have CV’d [Generate Information].
If the face is found to be deformed with chipping off, then the generation needs to be adjusted upwardsWidth and height of the imageBecause every picture has a size. Because every picture has a size, not enough size and need too many elements will be easy to squeeze together and cause the picture to collapse.
3.3 ControlNet
ControlNet is a neural network concept, which is to control a large pre-trained model with additional inputs.The ControlNet plugin in SD allows precise control of image generation through the model, e.g., uploading line drawings to make theAI control of character movement gestures, edge detection, soft edges and graffiti scribblesAnd so on.
Below is an example of a doodle scribble (first you need to upload a ControlNet reference image):
Once uploaded, select Scribble’s Control Type and click on the explosion icon.
Fill in the cue words (imaginative) : [A hot air balloon is floating in the sky and the full appearance of the balloon can be seen.], then click [Generate image]
IV. Summary
So far, the super full-blooded AI drawing tutorials come (online Stable Diffusion a key to use) to complete the introduction, the subsequent output will be more AI drawing-related chapters; access to the process if you encounter problems welcome to leave a message or private message exchange.