OlstFlow

The UX Dilemma: Evolution vs. Habit 🧠 AI models have evolved. Why haven't our interfaces? Professionals are already outgrowing the linear "Wall of Text" — it’s too limiting for complex tasks. But here is the real question: Is the average user ready for Spatial UI, or is simplicity still the king? My demo for the Gemini 3 Sprint 👇 (What do you think?) #GeminiSprint #BuildInPublic #UX #IndieDev #GoogleAI #SpatialUI #hackathon #gemini3 #GoogleAIStudio

Some people love AI. Some reject it. Some are afraid of it. I just keep doing my job — now faster, easier, and with more creative freedom.

POV: i'm a software engineer in 2026

Some words I want to share about the LichtFeld Studio philosophy and the role of the Python plugin system: With the growing adoption and improving capabilities of LLMs, my vision for the Python integration is to empower everybody to easily build custom tools. That's why LichtFeld Studio is transforming into a very fast C++ core with a Python layer on top. My idea and work in progress is to create a platform that is self-contained and can just be run without any installation issues. You can simply pull the latest nightly build onto your system and run it either from hard disk or even from USB. The bundled Python environment shall make it a breeze to hook up any advanced coding LLM to create custom tools for your workflow. This is currently in development for the next release v0.5.0 and works already to a large extent. This includes but is not limited to: • Mesh to splat • COLMAP integration • SAM3 integration • Diffusion models applied on top of the Gaussian Splat • Faster experimentation with training • Instant densification • And more… You can hook into most parts of the GUI already. The only limit is your imagination and you should be able to create everything your creative workflow requires. You can influence the development and you also own the tool. No privacy concerns, everything can run locally or remotely as you like. I am happy about code support but also about testing, bug reports, and most importantly financial support. If this project is useful to you, please help to make it sustainable by supporting this project with donations. Some options are listed on lichtfeld.io but also feel free to reach out to me to help keep this project going.

@NoContextHumans You also forgot about the a dual-use WC and the Restroom for dissatisfied women

Best bathroom sign I’ve ever seen

DepthMachina is out! 🎉 Blender add-on for zero-shot Mesh & Point Cloud generation using Depth Anything 3 + MoGe-2. Turn images into quick 3D bases (batch supported). Tested on RTX 2060 6GB. GitHub releases: github.com/OlstFlow/Depth… Patreon: patreon.com/posts/150086407 #Blender #BlenderAddon #AI3D #DepthAnything3 #MoGe2

Although this is nothing new, I still can't stop admiring the fact that we can create 3D images from a single photo with just one click. (Made with DepthMachina Blender add-on)

@Madisonkanna The hardest thing, when a machine takes the kind of work that was always machine-work, is remembering that you aren’t a machine. You’re a person. And what’s left — what matters — is the human work: feeling, loving, finding joy, showing compassion, dreaming, and caring.

as a software engineer, i feel a real loss of identity right now. for a long time i defined myself in part by the act of writing code. the pride in a hard-earned solution was part of who i was. now i watch AI accomplish in seconds what took me hours. i find myself caught between relief and mourning, awe and anxiety. the craft that shaped me is suddenly eclipsed by a machine. who am i now?

Damn, I felt it in my chest

@SteveWarnerFL Cool! The smoke probably isn't the best I've seen, but the liquids and light trails look great!

This is amazing. It would be extremely difficult to get these effects with prompting alone.

@shiri_shh It's so familiar 😅

me having 47 creative new ideas at 3 am instead of sleeping

SAM-3D Body works perfectly on 6 GB VRAM (RTX 2060) in pinokio.co ✅ This scene was processed in ~6 minutes. Good news for those who were disappointed by the “minimum requirements”: Pinokio lists 12 GB as the minimum, but this is (to many people’s relief) not true — I verified it 6 GB handles this task without any issues 💪 So if you were wondering whether your 6 GB GPU can handle generation — the answer is yes!

InfiniDepth: Fine-grained depth estimation at any resolution; models depth as a neural implicit field instead of a grid; - tops DepthAnything; - recovers sharp details w/ a lightweight 15M decoder. zju3dv.github.io/InfiniDepth/

I’m working on a Blender add-on called DepthMachina — generating meshes and point clouds from a single photo using MoGe 2 and Depth Anything 3. The idea: quickly get usable 3D geometry directly inside Blender for fast scene building. Useful for: architects / archviz, researchers, concept artists, environment artists. #Blender #BlenderAddons #3D #DepthAnything #PointCloud #ArchViz #AItools #3DWorkflow #ComputerVision #DA3 #MoGe

@emmanuel_2m @IsThisA3DModel I feel like human judgment is, in 99% of cases, an extremely inaccurate thing. In my life experience, people usually either underestimate or overestimate, and only a few turn out to be right in such predictions.

In 2026, I want to hear about all the haters that have screamed for years that AI will never ever be able to generate good 3D. 🫶🏻 (We're looking at you, @IsThisA3DModel & friends)

BTW: If you know any other great MIT / Apache-licensed models (like MoGe2 or DepthAnything3) — let me know. I’d be happy to consider integrating them into DepthMachina. Important: please don’t suggest research-only models (e.g., Apple’s SHARP) — those won’t be integrated due to licensing restrictions.

The Year of the Snake is coming to an end — a year that was anything but simple, yet deeply interesting. A year of struggle and uncertainty, of the old world collapsing and a new one taking shape. So much fear and curiosity. Nostalgia for what was — and anticipation for what’s next. One thing I know for sure: this year many of us became even more cyborg-like. AI-assisted thinking at the “periphery” has become an inseparable part of everyday life. I can feel my brain and the way I think has changed — irreversibly, forever. I hope for many of us this is a path of growth, not degradation. Happy New 2026 — I have a feeling it’ll be even more interesting.

Yes, thanks — I’ve already run Trellis without any issues. My main interest right now is specifically to try Trellis 2 and see how much the results have progressed compared to version 1, and what can realistically be achieved. For example, I managed to optimize Hunyuan 2.1 quite well by using Flash VDM, so even when requirements look high, meaningful optimizations are often possible.

@OlstFlow There's an older version of Trellis that you can try also. It's about a year old but it worked in my 8GB 1080 card without issues. Still does a great job and I use it in project work all the time.

Even more tests with Trellis2. The Refiner nodes in the VisualBruno repo do a remarkable job of adding detail and fixing problems with your mesh. github.com/visualbruno/Co…

@SteveWarnerFL Thank you for the reply. I wasn’t expecting a high resolution anyway — I was mostly curious whether it was possible at all. Now I know for sure that it’s at least worth trying.

@OlstFlow It's possible but you have to set your expectations properly. Trellis has three resolution presets. 512, 1024 & 1536. With 24GB VRAM I can run 1024 res. Sometimes 1536. But 1536 often runs out of ram. At 6GB you may be able to run 512 but likely will need at least 8GB.

AI lets you generate your own Santa for any mood 🎅✨ Prompt: Ultra-detailed 3D render, marvel-comic realism, dramatic studio lighting, sharp focus, high detail materials. Muscular Santa Claus in a tight red leather superhero suit (chest fully covered), torn sleeveless top with rugged Wolverine-like vibe, strong heroic silhouette, angry intense face (classic Santa face, no cyber implants on the face), thick white beard and brows. Santa is in a low superhero crouch (Spider-Man-like squat), one hand braced on the front knee (the near hand), dynamic tension. Right arm carrying sack pose (anatomically correct): right fist gripping the sack opening right next to his cheek/temple, sack resting on the right shoulder, right elbow pointing forward, upper arm in front of the chest, shoulder flexed forward, elbow bent ~90 degrees, forearm rising toward the head, realistic wrist and muscle tension. Environment: Santa crouches on the top of a huge defeated rusty robot head, only the upper crushed/deformed armor plate visible, robot eyes dark/off, metal is cracked, scratched, oxidized, battle-worn. On a fragment of the robot’s plating there is a stamped/painted date: “2025” (clear, readable). Behind Santa: a chaotic stack of oversized gift boxes and big crates, wrapped in colorful paper with ribbons, taller than Santa, filling the background. Details: cyberpunk boots and kneepads (tactical, futuristic), subtle cyberpunk accents on gear (gauntlet/boots), reindeer tattoo on the shoulder in a modern stylish graphic design. Background: simple dark studio / neutral backdrop, no city, no snow, no particles. Cinematic composition, slightly zoomed-out so the full pose and robot plate are visible. NEGATIVE PROMPT snow, snowfall, city skyline, neon city buildings, cyber face implants, goggles, mask, glowing eyes on robot, cute/cartoon style, low-res, blur, bad anatomy, dislocated shoulder, arm behind back, elbow pointing backward, broken forearm, twisted wrist, extra fingers, missing fingers, deformed hand, messy face, cross-eye, text artifacts, unreadable “2025”, cropped head, cropped sack