Convey

Convey is a product in R&D. It will use game technology & AI to easily create assets like videos & interactive experiences to explain concepts, code, & products

[Convey @convey_it]

Physically accurate audio-rendering is crazy computationally expensive, hacks are the way to go. Procedural techniques abound though, that's what wwise et al get you. I haven't messed with Unity's sound system much but I know Unreal now has some fancy procedural sound generation tooling.

[Convey @convey_it]

@gadirom_ Very cool!

[Convey @convey_it]

I was hoping to build a water colors style diagramming tool using p5.brush, but it's just too slow.

The effect is gorgeous and our tools should be more beautiful, so that's sad. I'm sure there are more optimal alternatives.

[Convey @convey_it]

@acamposuribe @seflless Honestly, not sure besides switching to using rect draw calls and playing with brush parameters.

I’ll be happy to dig deeper into the lower level algo and rendering if I go further and make a useful prototype, this approach is what I might check out

[Francois Laberge @seflless]

The natural next steps for people who really wanted this to be a real editor. Figure out how to replace @tldraw fill/strokes rendering with this, but leave all UI affordances for editing as is otherwise.

[Convey @convey_it]

@barany6m 👋

[Francois Laberge @seflless]

Prototyped a full web editor that generates real-time custom styled avatars. Supports randomly generating avatar ideas to pick from and editing the prompt and image influence.
🦸‍♀️🧙‍♂️👾
🐲🤖👻
🦹‍♂️👨‍🎨🦸‍♀️

Uses @fal_ai_data for real-time generative images.

The final detail to prove out the concept is to make it controllable at the part level and associate text descriptions for each part and automatically generate the prompt text.

[Convey @convey_it]

@alexanderchen @GrantCuster @Replit Love this exploration!

[Convey @convey_it]

I was gonna ask, I’m very interested in how you approached making yoga in general. How much was referencing browser implementations and how much reverse engineering and yes how did you test it as you went.

Related: has anyone tried to make tooling to compile only subsets of functionality of Yoga to make smaller builds?

[Convey @convey_it]

@tuhin Do you mean like @excalidraw or @tldraw ? Both support mobile and are open source.

[Convey @convey_it]

@disconcision I missed the context. Has there been some recent progress that stands out?

[Convey @convey_it]

No source code provided or live demo, only this hint at their approach:

"My implementation uses a framebuffer ping-pong to iteratively displace the input based on a noise value."

Source: https://t.co/Y3d9dm5qYX

My guess is that they meant they are basically making a GPU based water paint simulation using render textures for each time step, alternating between drawing to one and reading from the other, switching each frame (hence "ping-pong"). Water paint basically follows a reaction diffusion shader approach with optional model approximations of paper/canvas material and paint liquid.

Example video:
https://t.co/O923XLduGp

Example shader:
https://t.co/uUN2G5RLou

[Jesper Vos @jesper_vos]

@gheotil My implementation uses a framebuffer ping-pong to iteratively displace the input based on a noise value.

[Convey @convey_it]

I did spend some time looking for prio-art of hard real-time water color painting effects.

The most interesting things I came across was this real time animating demo of water color drawn fonts:
https://t.co/AN64bKaf33

[Jesper Vos @jesper_vos]

Watercolor/paint bleed shader study.
#webgl #glsl

[Convey @convey_it]

@jakedowns So you’d what have water paint type textures that you combine in this same way?

[Convey @convey_it]

My @seflless account was finally unsuspended! I'll be posting from there again moving forward. If you only follow me on this account, hope you join me on my main account.

[Convey @convey_it]

You had mentioned being super into it a while ago and I tried it, and had a meh experience. Tried again today using it to research existing projects doing real time water color effects and wow did it deliver. I happened to have used Google a lot trying to do this the classic way. It out performed me and found a project in less than 10 seconds that looks great.

[Convey @convey_it]

@sreekotay Now that you mention it, the example I forked is even more so:
https://t.co/1kstI3Oi10

[Convey @convey_it]

@emmanuel_2m 🤘

[Convey @convey_it]

@mkualquiera @emmanuel_2m has anyone used Scenario to develop infographic type visuals?

[Convey @convey_it]

@0xPHBD @acamposuribe I think you also could just generate image graphics once or on any scale changes in an editor, translation and rotation could reuse the graphics. Just the paint mixing/compositing is probably possible to do more cheaply on changes

[Convey @convey_it]

@0xPHBD On a quick search on most approaches they involve tons of shader passes and such.

@acamposuribe have you written up an explanation of your approach? It might be easy to improve at a glance by someone predisposed to WebGL rendering optimization.

[Convey @convey_it]

Yeah exactly. I think given you know the semantics and objects in your input image you could also inject automatic scene description into prompts. Like “a diagram with 3 shapes. One green diamond, a blue rect, a yellow rect, and an arrow between the yellow and blue rect. <user supplied text prompt for styling >”.

I made that connection when having similar issues with this avatar experiment: https://t.co/mMTXlJfsEi

[Convey @convey_it]

Avatar-to-baby generator. Continues my exploration of Avatar editors as a modality for generative AI.

You can easily retrofit any avatar system to make it specialized to another avatar type, here that's a baby head generator.

Really what I'm getting at is that configurator type UIs are super easy to use (and code), so go make one!