GPT Applications - Images and Video – Personal Science Guide to AI

Images from a Base Image

Tencent ARC PhotoMaker will generate photos from a base image:

All code is available on Github

Facebook

Imagine.Meta.com

prompt: “neoclassical style of the back of a boy scientist in a white lab coat facing an excited mob”

Alibaba’s Animate Anyone Github can animate any image to move however you’d like.

Magnific can upscale and remake a blurry image into something more useable. > Pro plan costs $39/mo, the Premium plan $99/mo and the Business plan $299/mo. When you opt for an annual subscription, you get two months free. You can cancel at any time.

Getty Images offers an image generator trained on their own database of images (to eliminate copyright risk)

NightCafe with links to multiple image generators.

Stable Diffusion XL

Interactive Image Generation

Decohere generates the image in real time as you prompt interactively.

Decohere: “neoclassical image of a boy scientist facing a stormy ocean. A firebreathing dragon peers over the horizon

Stable-Doodle lets you draw simple stick diagrams that it converts into full-blown art.

Transform a logo into a “stunning piece of art” with AiLogoArt.com

Ideogram.ai : generate images with realistic typography.

Aragon.ai turns selfies into professional headshots for $29.

Prompt: “super Cute puppy dog holding a flag that says”Go Leah!“. cinematic 3d render”

example “Richard Sprague statue in the style of Rodin’s thinker. Somebody threw a baseball cap large red label “Mercer Island” on the statue’s head. Photorealistic.”

“Richard Sprague statue in the style of Rodin’s thinker. Somebody threw a baseball cap large red label “Mercer Island” on the statue’s head. Photorealistic.”

Modify a Base Image

FreePik will generate up to five images for free ($12/month for Premium) but I don’t find the quality compelling at all. Now incorporates Magnific, the image upscaler.

It converted this:

into this

Alphabetic Characters

Make your own custom alphabet characters using Google Lab’s Gentype

Star constellation example of Gentype. Prompt “star constellation map, in the night sky, above treeline, nature photo”

Video

Make short videos for free with (Chinese app) Kling: @kling_ai

(source: Factorial Funds)

Pika Labs makes short video on demand, including sound effects.

Luma Labs generates short video clips for free

Luma Labs. Prompt: “realistic dolly shot of chickens running through a modern office with cubicles, modern furniture.”

From China: see - Kling by KWAI is throwing hands with OpenAI’s Sora. It creates 2-minute long videos with impressive consistency.

Thoughts about Sora

A technical deep-dive explains how it works that estimates it was built with 4,211 - 10,528 Nvidia H100s for 1 month. Extrapolating to what would happen if Sora gained significant market share on TikTok and YouTube, that’s on the order of 720K H100s.

2024-02-24 2:18 PM

I haven’t studied the details of the new video generation model from OpenAI but I’m pessimistic about why this is a major advance.

Seems like a more effective way to build realistic videos would be to simply issue commands to a standard game rendering engine. Set up the characters and backgrounds and programmatically tell it to move in pre-specified ways.

But some of the results are of course incredible