Leonardo AI – step-by-step guide

Detailed explanation in Mandarin (Taiwan) - very good one!

  • 150 free tokens per day

If you have no idea how to write a prompt (1:05)

  • hit “Prompt Generation”
  • enter a basic prompt idea
  • hit “ideate”

Guidance Scale (1:45)

  • the higher the number, the more guided it is
  • low number = more creative
  • best 7-10 
    • too high or too low = weird images

Step Count (1:55)

  • the higher the number, the more detailed the image is
  • but also takes more time to generate and costs more credits 

Tiling (2:00)

  • seamless patterns (ideally)
  • prompt: 4-sided joint

More background / zoom out (2:55)

  • Unzoomed image
    • move object further away 
    • toolbar under an image

Extend background (5:40)

  • Edit in canva (toolbar below an image)
  • You can upload your own image or photo (real or fake) or use what the tool has generated (6:20-7:10)
  • click arrow to see more images for selection

Remove background (3:05)

  • middle icon on toolbar under an image

How to enlarge or add details to an image (3:30)

You will see a purple dot at top right corner of the image so you know it has been upscaled

  1. Creative upscale (3:30)
    • details of original image may be changed
    • toolbar under an image
    • e.g. 800 x 1024 vs 2112 x 2688 
  2. Creative upscale alternate (3:45)
    • more details of the original image will be kept (look more like the original image)
    • toolbar under an image
    • e.g. 800 x 1024 vs 2112 x 2688 

Use image as reference instead of writing prompt (4:00)

  • Image to Image
    • toolbar under an image or on the left 
  • Init Strength
    • the higher the number, the more the newly generated image will look like the original
    • lower number = more creative
  • you can also change the prompt on top now if you want 

Negative prompt for what you want to remove or don’t want to see in the generated images (4:20)

Use “Prompt Magic” if you don’t like what you see (4:55)

  • Prompt Magic makes the software generate images that closely match your prompt (i.e. less creativity)

Leonardo Style (5:05)

  • adjust the style 

Make a character do a certain pose (5:20)

  • ControlNet > “Pose to image” (left menu)
  • upload an image with someone in that pose

Image touch up (7:10)

  • remove an object (7:15)
    • turn on “mask only” 
    • add a prompt to fill up the missing piece
  • add an object in a certain part of an image
    • draw mask (7:30)

Download an image (7:50)

  • left menu

Create a brand collection / same characters for a book (8:05)

  • Training & datasets (left)
  • create new dataset
  • give the new dataset a name > create dataset
  • upload 8 – 15 images to train the AI
  • hit “Train model” > model description > Instant prompt (keywords?) > “start training”
  • it will take about 30 mins, then you will get an email 
  • finetuned models (left) > your models (top) > view > generate with this model 
  • Type a prompt (e.g. creamy mami 小忌廉) > generate (8:45)

When an image is distorted (6:55)

  • when your face looks weird
    • change image dimension
    • lower guidance scale  (more like the original photo)

Select a cover photo of your collection (10:25)

Select a cover photo of your collection (10:25)

  • hit purple button “use this image as the cover of (collection) model” 

How to get to bottom of page? (12:00)

  • hit “end” button twice

How to quickly generate a lot of images similar to those created by other creators? (12:10)

  • click on the image
  • hit “image2image” (next to image)
  • hit “generate” (right)

How to return / go back to make changes (18:35)

    • hit “cancel” or “ctrl + z”

ControlNet to control Leonardo to follow certain things in an original image (2:50)

  1. Pose to image
    • keep the post of the character 
  2. Edge to image
    • keep the composition as the original image
  3. Depth to image
    • layering

This video focuses on print on demand products

  • Better than Midjourney and BlueWillow
    • can take prompt samples from other people
    • can do it on a browser
      • no need to join the Discord server (but you can)
  1. signing up ()0:30)
  2. learn the prompts other people are using (1:10)
  3. go to Finetuned Models > Community Models > Type what you are looking for in the search box (2:20)
  4. go to Category to select Texture
  5. you can follow (bookmark) / download / copy prompt (3:10)
  6. Go to AI Image Generation (left menu) > paste your prompt and make changes (3:25)
  7. select Leonardo Diffusion (3:40)
  8. you can choose image dimensions etc on the left menu (3:50)
  9. Results you will get from the same prompts will always be different! (4:20)
    • you won’t repeat someone else’s pattern
    • but you also won’t get the same image again!
  10. Download and sell your artwork (4:50)
  11. How to do it on rebubble (5:45)

Tiling for repeated patterns to sell digital art to make money (4:15)

  • prompt: 4-sided joint

Get higher quality images to make money (3:05)

  • Prompt Magic
    • get better results when you run out of things to try
  • HD upscaled Image
    • more detailed image
    • 5 credits 

Consistent character for a book or a brand (4:40)

  • give the character a name