nuclearsugar

Position: Motion Designer

Location: Austin, TX

Remote: Yes

Willing to relocate: No

Technologies: Maya/Redshift, After Effects, StyleGAN2+3, AnimateDiff, Stable Diffusion

Portfolio: https://www.jasonfletcher.info/

Position: Motion Designer

Location: Austin, Texas

Remote: Yes

Willing to relocate: No

Technologies: Maya, Blender, Unreal Engine, Redshift, After Effects, Premiere Pro, Photoshop, StyleGAN2+3, AnimateDiff, Stable Diffusion

Portfolio: https://www.jasonfletcher.info/

Linkedin: https://www.linkedin.com/in/hey-jason/

But in reality it's more subtle than that. Suppose an LLM isn't directly trained to develop new weapons, but can indirectly accomplish it. This collaboration opens that door.

Is the helium cooling aspect part of a closed system?

Very excited to play with this. Some of my latest experiments - https://www.jasonfletcher.info/vjloops/

Thanks! Indeed the cleanup required for the 'Nature Artificial' dataset was intense and forced me to consider other techniques. It was around that time that I realized that I could use Stable Diffusion to output 10,000 to 50,000 images, which has proved to be a useful dataset creation tool for training StyleGAN2 or StyleGAN3. Yet nailing down a SD text prompt that will continuously output images in a specific style and context can be challenging.

StyleGAN2/3 is quite sensitive to fine-tuning models with a dataset that contains lots of variation, such as the Pokemon dataset. But it will converge much better with a dataset of just Pikachu images in tons of different poses. It seemingly wants to find the most common pattern in the dataset and interpolate within that space. From there nailing down the ideal value for the gamma attribute is difficult, but luckily can be reduced every 1000 to 2000kimg until the model no longer trains favorably. Currently my typical gamma strategy is: 80 > 10 > 5 > 2

On the topic of exploring the innards of machine learning models, here are some visualizations I created of the hidden internal representation layers within the StyleGAN3 FFHQ1024 model.

https://youtu.be/2aDeS_RFqHs

https://www.jasonfletcher.info/vjloops/index.html#internal-r...

Anyone in the EU experiencing undeleted stuff?

Is it possibly they are specifically geotargeting IP's to undelete stuff and therefore avoid GDPR?

I use Google Domains frequently because they support domain forwarding for both HTTP and HTTPS requests. This is critical for end users that type or click an HTTP address.

Can anyone suggest a domain registrar that supports forwarding over SSL?

I've been using a HostingBy.Design seedbox (formerly Seedbox.io) for years to distribute content to my patrons for 3 years. They have excellent uptime and their customer service is knowledgeable.

Stable Diffusion trains StyleGAN2 - https://www.jasonfletcher.info/vjloops/

Stable Diffusion paired with StyleGAN2 makes for some fun results - https://www.jasonfletcher.info/vjloops

I've been experimenting with creating image datasets via Stable Diffusion, then training StyleGAN2, and compositing in After Effects. It's a new mode of art creation for me that I cannot recreate otherwise. https://www.jasonfletcher.info/vjloops/

Thrilled to hear that you've found my Cream of the Crop pack useful! So rarely do I get to hear back from performers.

Not sure if you also saw my mashups pack, you might enjoy it - https://thefulldomeblog.com/2020/08/18/nestdrop-presets-coll...

You might dig NestDrop for queuing up presets and then outputting the visuals to a Spout video stream, which most VJ softwares can ingest. Or you can install the OBS Spout plugin into OBS and then record the full rez visuals directly. Here is a tutorial I made demoing it - https://youtu.be/yHqs-EeoIfQ?t=414

This also allows you to inject live video into the Milkdrop engine. Here is another tutorial I made - https://vimeo.com/599028390 https://vimeo.com/450122956

I did a ton of curation of presets and here my cream of the crop pack containing 9,795 presets - https://thefulldomeblog.com/2020/02/21/nestdrop-presets-coll...

More info about NestDrop - https://nestimmersion.ca/nestdrop.php

And on a tangent, you might enjoy my StyleGAN VJ loops - https://www.jasonfletcher.info/vjloops/

Currently a high end GPU is required to avoid long render times.

Topaz Labs Video Enhance AI is the best in class commercial AI option. https://www.topazlabs.com/video-enhance-ai

ESRGAN 4 Video is probably the best open source option. https://colab.research.google.com/github/MSFTserver/AI-Colab...

A good list can be found here - https://pharmapsychotic.com/tools.html

Maya or Blender batch rendering

Thanks for this, exactly what I was looking for

Incredible! Thanks for sharing this.

I recently upgraded MacOS and now TextEdit occasionally crashes randomly. And pasted text has bizarre spacing even in plain text mode. Are they eating their own dog food?

For anyone interested in experimenting with an open source text-to-image AI tool, check out DiscoDiffusion on Google Colab - http://discodiffusion.com/

"Characterization of JWST science performance from commissioning" - 2022 July 12

https://arxiv.org/ftp/arxiv/papers/2207/2207.05632.pdf

Check out the chapter: 4.7 Micrometeoroids

DallE2 + Topaz Gigapixel AI works amazingly well.

FYI, each text prompt submission uses 1 credit and then renders out 4 images.

An interesting use-case is to inject a webcam feed into the Milkdrop engine (alpha channel included) https://vimeo.com/599028390

Also useful for recording the visuals of a VJ performance at full resolution without needing to sacrifice a whole monitor.

But typically Spout is used to share a video steam directly on the GPU between apps, such as Resolume and MadMapper.

OBS is quite useful for recording Spout or NDI video streams. I use these two OBS plugins often - https://github.com/Off-World-Live/obs-spout2-plugin https://github.com/Palakis/obs-ndi

At first I thought this was an elaborate SCP story taking over Wikidot.

IMO read the news about twice per week. But never watch the news on TV.

Hmm that's an interesting idea.

After Effects still does not support exporting to the MKV container and so it doesn't fit into my typical rendering workflow. Hence I've stuck with MOV using HAP-alpha.

As a experiment, I rendered out a PNG frame sequence (with alpha) and then used FFMPEG to render out both an MOV and MKV using the hap_alpha codec.

--- When I tried importing both videos into Resolume then the MOV was successful but the MKV was rejected. Doesn't seem to be a supported.

--- VLC could not playback the MKV.

--- MPC-HC successfully played back the MKV and reported it was an rgba video.

So it was interesting experiment, but not something that my necessary software pipeline supports.

Here is some of my 3D animations for context. https://www.jasonfletcher.info/vjloops/

Resolume, VDMX, Modul8, TouchDesigner, MadMapper are the most common softwares that performers rely on