WHEE usage tutorial

WHEE usage tutorial: 1. Upload the image, and select the model and corresponding creative description; 2. Choose the style model you need, and you can freely match a variety of LoRA overlays; 3. Add and add based on changing the style Referring to the similar types of pictures, the control models can be superimposed and used; 4. Fill in the editing effect name and introduction information, and then publish it.

WHEE has just been launched. Maybe you don’t know how to use it yet. Here is a tutorial on how to use WHEE.

Usage tutorial:

First upload a picture

##1. Model creative description

1) Select a basic model to set the overall style tone for the effect. For example, select the model "Ancient Poster" with reference to the picture

Only one type can be selected and cannot be superimposed

2)Supported Chinese and English descriptions, description content suggestions: conceptual, large-scale, stylized keywords are written first, keywords describing the content of the picture are second, and finally keywords describing details; for example (picture quality prompt words), (Screen theme content) (Style), (Related artist), (Other details)

3) You can use brackets to modify the weight of the prompt word (the default weight is 1), For example, (redhair:1.5) increases the weight of this keyword by 1.5 times; by increasing the weight of "redhair", the proportion of keywords gradually becomes higher, and the proportion of red in the screen also increases. When the value reaches 2, the screen will collapse. It is recommended that The maximum weight value does not exceed 2; on the contrary, setting the weight of the keyword to 0.25 will reduce the weight proportion of this keyword

2. Style model

Optional items: [Style Model LoRA] can add to the effect, and fuse the basic model to make the effect closer to the target style; you can freely mix and match multiple LoRA overlays to try to create different effect styles

3. Picture control

1) Picture control can add similarities to the reference picture on the basis of changing the style, such as action (gesture detection), color (color), environmental background ( Edge, depth, line detection), etc.

2) The control models can be superimposed and used together, but the more overlays, the longer it will take. It is recommended that adjustments be made within one control model as much as possible. For example, if you want this effect Try to get the character poses and clothing as close to the reference picture as possible, and add the "canny model" to generate

. When generating more style effects and want to select a certain style effect for parameter fine-tuning, You can find the "Edit Again" button in the picture to modify keywords and parameters.

4. Publish

After achieving satisfactory results, click on the picture to generate a large Picture, click on the upper right corner to download and publish (edit effect name and introduction information)

Creation Tutorial:

1. Interface

After entering the Meitu AI creation tool, enter your creative content on the left, set the style model and generation parameters, and you can generate creative pictures with one click on the right.

2. Creative description

The creative description content is the core content of the text and pictures. A concise word combination is needed to describe the creativity to be displayed, such as : "Chinese landscape painting, with towering mountains and shrouded clouds and mist." There is no need for complicated sentences. Try to enter short nouns and phrases, separated by commas, so that AI can better understand your thoughts. The overall input details are as follows:

Use more key rhetoric of professional art, such as the following general descriptions of the level of picture fineness: Unreal Engine, Octane Rendering, 3D Rendering, Masterpiece, Best Quality Star, High Resolution, etc., are some common picture modifications.

You can add () twice to the words that need to be emphasized to increase the weight, such as: "(bust), close-up of a doll with pink hair, a detailed painting, fantasy art, Gurwitz-style art The work "big watery eyes, beautiful pink little girl" is to emphasize that the picture is mainly a close-up of the body.

The descriptor can be entered in English. The first half describes the important scene structure, the middle part is the details of the characters and environment, and the second half can add a description of the artistic style of the picture, such as: intricate, agirlinthestreet,alley, (1girl), ( messybluehair:1.4),(beautifuldetailedglowingblueeyes:1.2),makeup,earrings,choker,masterpiece,bestquality,highres,realistic,cinematiclighting,lookingatviewer,(cyberpunk:1.4),(t-shirt),(cargopants),(coat),( opencoat),anime,beautiful

face,2.5D,unrealengine,octanerender,3drender,intricate,1 girl on the street,alley,(1 girl),(messy blue head

Hair), (Beautiful Detailed Glowing Blue Eyes), Makeup, Earrings, Choker, Masterpiece, Best Quality, High Resolution, Highly Realistic, Theatrical Lighting, Look at the Viewer, Cyberpunk, T-Shirt ,tights,jacket,open jacket,cartoon,beautiful face,2.5D,Unreal Engine,octane rendering,3D rendering

Unwanted content can be input to reduce distracting factors, such as only one scenery is required For beautiful pictures without people, you can write people, people and other nouns in the negative input box.

3. Model selection

Select the model of the corresponding style to generate relevant pictures according to the creative description. Large models can be combined with style models to create different painting styles.

4. Other basic parameters

Size and proportion: It is recommended to choose a smaller size when running the drawing at the beginning to reduce time-consuming, and wait until you have your own When you are satisfied with the style of the picture, increase the size. For portrait photos, you can choose vertical or square proportions, and for landscape photos, you can choose horizontal composition.

Creative Relevance: The higher the degree of adhering to the initiative description in the control image generation process, the more the image will fit the description content, but a higher degree will reduce the creativity of the AI ​​itself. The recommended value can be in the middle Low position.

Sampling steps and samplers: Different samplers have different time-consuming and effects. You can try different samplers multiple times to get the best results. For the first time, you can choose the default fast sampling EulerA. The higher the sampling step, the faster it will be affected, usually around 25 is enough.

Reference picture: Create a picture based on the picture you uploaded, and apply the creative style to this picture. When using this function, pay attention to whether the proportion setting is similar to the proportion of the reference picture, otherwise it will be cropped.

Screen control: Screen guidance can be carried out through the ControlNet model. For example, if you import a photo of a desired character posture, you can use the posture detection model to guide the final generated picture to match your posture vision. Or use deep models to control the spatial structure of the scene.

The above is the detailed content of WHEE usage tutorial. For more information, please follow other related articles on the PHP Chinese website!