Hello there! Flux generates text perfectly. But its style is not always interesting and beautiful. The best text generator at the moment is ideogram.

I collected several hundred images from the top sections of the site. Reduced them to 1024 size and trained flux lore on 6 epochs and now you can make almost the same beautiful text as in ideogram.

Small example of dataset

Caps were made with Florence-2.

Training resolution is 1024, 6 epochs, 10 iterations per image.

How to use

Use LoRa from 0.5 to 1.5, the higher the weight, the higher the creativity, but the readability of the text may deteriorate.

You must use hi-resfix. at upscale level 1.3 - 1.5 to reduce artifacts.

Some examples of prompts

ideogram1024 style, a woman standing and holding a neon sign that reads "Make it Happen, Shock Everyone" against a dark background.

ideogram1024 style, a poster with a pink background featuring a butterfly and flowers, as well as text "Laura". The butterfly is surrounded by a variety of colorful flowers and butterflies, creating a vibrant and cheerful atmosphere. The text on the poster is likely a name, adding a personal touch to the artwork.

ideogram1024 style, a poster with a picture of a rocket and the text "Believe in the power of positive thinking, it changes everything" written in bold, white lettering against a bright blue background. The rocket is depicted in a dynamic pose, with its wings spread wide and its tail pointing up, as if it is about to take off. The text is surrounded by a white border, giving the poster a modern and eye-catching look.

Best generation settings

I use Flux.1-Dev Hyper NF4 model that works on only 8 steps.

I will be grateful for feedback and publication of your results. Hugged🤗

I talk more about neural networks on my YouTube, Telegram and Boosty. I would be glad to receive your subscription and support.