Shawn Wilkinson

made a physics based dice game. it's kinda fun - dicerun.io

Is self-learning AI just much larger context windows, and an efficient subprocess to move information in/out to an external store?

@ZitaoTech Do you do any paid commission work? Love this idea.

16+512 setup cyberdeck monster - HackberryPi 5

@prodialabs ,one of the world's fastest image generators, uses Cloudflare R2 for AI model storage & transfer—calling it the "fastest and most economical solution". See how they built with Cloudflare 👇

@cocktailpeanut Would love to test

Starting Pinokio 4 Alpha Test Today. To apply, reply here or let me know on Discord.

Not sure how to feel about OpenAI Codex. It was able to increase test coverage on one of my projects, but Codex was not able to get Jest installed properly to run tests...

@cocktailpeanut How is the pinokio API for WAN? I could build this right now just by doing a multi-path video generation.

if there's a way to do this very efficiently (let's assume it takes 1 second to generate a future frame), we can: 1. Generate 100 different futures in 100 seconds 2. Pick the future I want to see happen 3. Do a start/end frame to video interpolation

How about instead of image-to-video, image-to-futureImage? Instead of having to generate the full video, only generate a SINGLE frame N seconds into the future that would be generated if you generated the video using img-to-video.

@cocktailpeanut Do you think this setup could work with more than 1 GPUs?

@super3 I think it was 68 minutes on a 4090 but uncertain if this is the correct number, trying again 😅

What's the longest video generated you've seen with mochi? how many seconds?

🚨New Model Drop 🚨 Mochi by @genmoai , the SOTA in open-source video generation is now available for everyone on Prodia with a Pro+ or Enterprise subscription Here are some samples of what you can generate the first one is the classic prompt “astronaut riding a horse on mars, 4k”

@cocktailpeanut How long did 7 seconds take you?

@super3 Already tried 7 but just curious about the top record x.com/cocktailpeanut…

the easiest way to add generative video to your app

Experimenting with #mochi from @genmoai using @prodialabs's API. One thing this model really seems to excel at is macro shots. An example:

@ajayj_ What is the current generation time for this?

Run Mochi 1 with one 4090 GPU. No H100 needed. The official ComfyUI nodes for Mochi 1 are out with fast and efficient community kernels.

📊 OKX Ventures Q3 Report: Staying ahead with a continued focus on major ecosystems like #Bitcoin, #Ethereum, and the expanding world of #AI. okx.com/learn/ventures… Discover our latest insights and strategic investments shaping the future of blockchain and tech innovation! #OKXVentures #Crypto

Today @storj announced Colby Winegar's promotion to CEO 👏 He previously served as CRO driving partnerships, tech alliances, acquisitions & new customers that have transformed the organization. Full press release here hubs.li/Q02WhgGm0 🌟🚀#CEO #cloudleadership

@Kijaidesign Super helpful data point. Have you tried running it across multiple 4090s? Let me know if you need credits or access to test.

@super3 Shorter clips are decently fast, around ~5 mins for 49 framese, 163 frames took 20 mins which is quite slow, but not all speed optimizations are used yet. Interestingly the motion itself can be seen as low as 4 steps, so potentially one could find good promp/seed lot faster

As bit of a personal challenge I decided to get the new Mochi (github.com/genmoai/models) -text2video model running locally on 4090, this was generated under 20GB VRAM in ComfyUI. It's not perfect as I had to use tiling to be able to decode the whole thing, github.com/kijai/ComfyUI-…

@TopperDEL I just ran the base model code via command line. No great tools yet. Working on it.

@super3 What Tool helped you create that one specifically?

Open source video models are getting way better. Here is my first attempt with Will Smith eating spaghetti with made with Mochi.