AI Image Generation

Everyone talks about GPT, but you can also generate images. Lately, I’ve been playing with Flux, which is an image generator. You can generate images of me using the idvorkin with this lora on replicate.

montage of the sequence

Fun Demo

  • Let’s start with the Mona Lisa for ChatGPT

“A portrait of a woman with an enigmatic smile, seated in front of a distant landscape. She has dark, flowing hair and is wearing a modest dark dress with a translucent veil. The background features a winding path and a bridge, with a serene and dreamy landscape under a soft, diffused light. The overall mood is mysterious and timeless.”

  • Let’s try that with a lora of the Mona Lisa

  • Now, let’s add me to that scene “A portrait of MNALSA with an enigmatic smile, seated in front of a distant landscape. She has dark, flowing hair and is wearing a modest dark dress with a translucent veil. She is sitting beside her boyfriend, a middle-aged bald man, idvorkin. The background features a winding path and a bridge, with a serene and dreamy landscape under a soft, diffused light. The overall mood is mysterious and timeless.”

Talking Head Videos From your Heard

Woah, this works surprisingly well:

https://youtube.com/shorts/Y_YxRPC46-E?feature=share

Igor Image Generation

In the prompt, I’m idvorkin,

Igor doing something

You can blend in:

  • Mona Lisa: huggingface.co/fofr/flux-mona-lisa

  • Scarlett Johansson:

  • Real and Boring - kudzueye/Boreal - Keyword is photo

Blend multi loras (equal weights)

Flux

As of April ‘24, Flux.Dev is the new open source hotness. You can train a LoRA on yourself, and then blend it with other LoRAs to do fun stuff. Here’s me playing with stuff.

History Lessons

Completion

How it works

Diffusion

LoRAs

Interesting - So the original model is like 23GB, a Lora seems to be 200MB, or 1/100th the size. You can load any concepts you want.

Blending LoRAs

Providing Inputs

Text 2 Image; Control Nets; In Painting