AIArtist Art Poster Composition Design Model SD1.5

AiARTiST Art Poster Composition Design Large Model ArtComposition version 1.5

What is AiARTiST?

Due to having a composition design assistant Lora, wanted to find a matching base model in the 1.5 large model

But felt that the styles were not quite suitable, decided to burn one, and then integrated that Lora into it,

Tested it and it turned out pretty good, decided to share it with everyone. As the saying goes, throwing out a brick to attract jade, let's progress together.

2024/1/8

The main issue with this model is... it needs to be high definition when magnified, otherwise the training materials get compressed and the image gets messy.

In the interface of the text-to-image model, tick "High-Resolution Restoration," also known as local SD for high-definition enlargement or HiRes,

Use the fastest algorithm "Latent" to enlarge the image by 1.5 times in latent space, and the generated image will look much better!

2024/1/8

Paired with the AiARTiST Composition Design Large Model, introducing the "LoRA First Release"!

AiARTiST-Live Room simple background generator LORA model version 1.5:

Direct link:  https://www.shakker.ai/modelinfo/e4c559c3f2d04a99bfeb3b15dc1b27a7

2024/1/10

Paired with the AiARTiST Composition Design Large Model, introducing the "LoRA Second Release"!

AiARTiST-Bookstore Cover Design Assistant LORA model version 1.5:

Direct link:  https://www.shakker.ai/modelinfo/1108bbb12a0646759b2b68d8c2ebf209

2024/1/12

Paired with the AiARTiST Composition Design Large Model, introducing the "LoRA Third Release"!

AiARTiST-Chinese Food Material LORA model version 1.5:

Direct link:    https://www.shakker.ai/modelinfo/bf2bb9a32b6243989939a85b787e6f6c

AiARTiST 0.9 What's New

The main purpose is to better utilize AI creativity, truly achieve design equity, and popularize the productivity of art creation.

Not particularly concerned with facial models, so the training set does not include specific portrait models (so there's no need to work on hands or similar).

It would be nice to pair it with a series of design assistant Loras that will be released later, to effectively reproduce the style of some previous works.

————————————————————————————————————————————

AiARTiST 0.9 What's New

This model is very niche, suitable for design, illustration, scene Lora, and general characters, 2D, 2.5D LoRA can all try generating images.

What makes it stand out is the assistance in poster design, which should work well with visual design, art, and fashion display LoRAs.

AiARTiST has better compatibility with traditional negative words like artist_name, signature, artist_logo, watermark,

You can quickly use these words to simulate the feeling of the final image and then use other design tools to complete the process.

Recommended Prompts tag list (for reference only. You don't need to use it when generating images; masters, please ignore):...

Neg_Prompts: Try to avoid negative words as much as possible, let AI's brain run free, save our thinking; laziness is the first productivity!

Q&A

Do you have a VAE built-in? Yes.

Is it meaningful to build such a large model? I don't know.

What is the appropriate resolution for generating images?

I usually use 384x768 or 512x1024 (other recommended ratios like 512x768 are also fine, or you can create landscape images), then enlarge by 1.5 times using Latent, with a redraw amplitude of 0.5. If it looks good, then send it to image enhancement for refining. Lastly, use high-definition post-processing plugins to ensure that the PNG image information is not lost. This resolution is mainly to adapt to the visual habits of smartphone screens. Other ratios are also acceptable. However, when stretching the training set of SD1.5 too much, the accuracy of the material is reduced.

How much CFG is suitable?

I don't think I am more creative or artistic than AI; with that premise, I try to keep the CFG low unless it is to lock the content of LORA. This is usually between 4 to 8. After lowering the CFG, just enjoy watching AI go wild on its own! For subsequent distillation algorithms like LCM/LCM2, 1.5 is sufficient, and you can use dynamic CFG.

What sampling method and enlargement algorithm should be used? This varies from person to person as everyone has a different style, don't we?

Other questions await your feedback.

——————————————————————————————————————————

If there are any issues, please leave a message on the Lib model page; for business cooperation and V contact AiARTiST | Metaverse Digital Person/AiGC/ Metaverse direction.

"More fun uses, visit the model author's homepage to get added to the V group"