Mads Hagbarth Damsbo

@key_vfx @YouTube @YouTubeCreators @andrewpprice @JonathanWinbush @SirWrender @badxstudio 🤨, hmm Nuke in LA could out of context maybe get flagged by a silly automated system. Can’t see what else would cause it. But that is like years ago.

@CompAcademyVFX It was released a few years ago, tho invite/request only. I have yet to stress test it as the tracing is still being worked on, but with enough vram you should be able to go to a decent size. And looks awesome, there is a lot of cool cloud tech from the gaming scene to tap into.

Awesome. Playback speed looks great too, I wonder how big of a VDB you can load in? Also did you ever release Mist by the way? I remember seeing it a few years back and was always hoping for that to be available. I've been messing around vibe coding sort of a volumetric noise thing, but it's more for shaping specific clouds / wispies with different erosion controls

Had a blast playing around with #EmberGen during the holidays, so decided to make a little bridge to Nuke (and Fusion via OFX), to get some quick volumes into Nuke.

@RohanCreates Neat😍

Another COPs experiment #houdini

@daxcity @LunarSoftware Such an amazing bite-sized journey. Audio and Visual stimuli. 🦾🤖🦿

🤖 #routine @LunarSoftware

@seevua @andreasgysin Thats neat

光ファイバーをゆらゆらさせて 文字や記号を表示させてる #NT東京

@titomaldegorf It is still on the drawing board (pun intended)

@xads What app!!! 🥰🥰🥰🥰🥰🥰

Doodling around with my drawing app. Not sure where this is going, but with a bit of work it could be interesting.

@misaki_mofujp 😰

shaderでどうやってビジュアルを作るかを聞かれたとき、数式を作って座標をコントロールすることによって欲しい色や形を作るんだよ言うとshaderに馴染みのない人にはあんまり理解されないことが多い、でもそれ以上の説明が難しい。

@artixels haha, with Wan2.2 coming in 2 weeks and 2.5/3 later this year i think that is a uphill battle😅

@xads Of course it is !! I will remind you when I feel you are done with WAN 😅

artistic exploration via sampling a tiny neural network's memory. curate your own data and train your own model. #diffusion #neural #generative

@artixels One of those that is somehow allready in my bookmarks, but forgotten all about it 🫣 But it's quite fun to play around with i assume.

@xads Graphics guy diffusion model: github.com/tchambon/IADB You don’t need the attention layers though

A few more tests, this time from the back yard combining some basic Nuke masking with mask sampling and inpt. The idea is to provide a basic growth mask as a guide, then the mask sampler expands the mask to the size of the projected latent output giving a bit of wriggle room.

I was doing a demo at Midas on VACE and was asked if we could make a thing burn, and it failed miserably😅. One thing is to make a thing burn, the other is how do you control placement, flow and look of that fire. So spend a few late nights tinkering...

@Aurelien_Gz I bought a ps3 just to Play flOw and Flower… then journey came… favorite console generation 🥰

this is a masterpiece.. flOw, my favorite flash game ever.. made in 2006 by some wizards I've spent hours lost in this peaceful world..

Haha, yea. But exactly, or just damaged film in general. I guess my point is if you can do it analytically, then a model can do it as well. At my previous employer NordiskFilm they had a film restoration department that did nothing but scanning old film, dust and grain busting and then onto the grade. In VFX we did a few touchups for things that were too damaged mostly inventing stuff that was not there to begin with like you say, but trying to be faithful to the original material as possible. At the end of the day it is like game remasters, if you can keep it faithful (and consistent) to the original material then why not? (We need to fix eye-gaze tho, it seem like every model struggles with this, even with guides.)

Yeah. I've had 100+ comments on LinkedIn saying to me "but can't you see that it's not the same thing!" :D (Like I wouldn't have noticed it. :) I do think that colorization of film is kind of solved problem by now...? That has been done for a long long time. Sure one can probably make it more effective in terms of determining accurate colors by polling or cross-referencing the colors given by traditional algorithms with colors one gains from here, tweaking them in those directions. The slippery slope comes from - "ok so let's add some detail to the blown out parts of the original film" - so how far does one take this. Add texture to a jacket - add a pocket to a jacket - change its design.

Magnific 'Structure Reference' + RunwayML "Restyle video" to colorize old movies ... I think one might run this in a way you use this to help to determine the color layer and then apply the colors on top of the original black and white image... ... would work best if you would do a separate init frame for each cut...

@guycalledfrank Ohh yes works the other way around as well☺️, I remember that time when a Skype image appeared inside Nuke when reading from a uninitialized buffer (opencl). It’s actually an interesting way of beaming data from one process to another over the GPU.

FUN FACT: it seems In CUDA you can overwrite completely unrelated data by writing to a buffer out-of-bounds. AFAIK drivers guard against it for regular GAPI shaders, but here... just a fixed a bug that had an out-of-bounds write to a byte buffer corrupt an unrelated uint buffer😳

@maksvanleeuwen Congrats with the release Max! Looks solid!!

Go check out Pixel Repeat! It does cool stuff like this:

@guilanier neat!

Throwback to one of my favorite projects from last year! Wanted to revisit this text line drawing made with canvas2D techniques — guilanier.github.io/anoyo/line-type #javascript #interactive