W. S. Waldo retweeted
SolidCraft 0.45 is out! It has DRAWINGS! 📐
solidcraft.io
- Drawings: sheets, views, dimensions and notes that work like a sketch
- Base and projected views, parts lists, and SVG or PNG export
- Sheet metal library of materials, gauges and rule packs
- Pattern whole assemblies, and big assemblies open faster
- Saving never refuses, and a driver crash at start-up is skipped next time
Wait... What?
提示词写得再细,也只是在许愿;真正的导演工作,发生在片场——调度、走位、运镜,全是实时发生的决定。
Pika Enact 把“在片场导戏”做进了 AI 视频:拖动演员重新调度,开口跟演员讲戏、喊 Action 开拍,鼠标实时控镜。生成不再是开盲盒,而是坐回导演的位置。
pika.art/blog/pika-enact?utm…
1
27
Comfy Cloud is now a viable option for teams. Push your custom models, Loras and workflows to the cloud. The whole team has access. And the GPU options have increased. This is what we've been waiting for.
Your custom ComfyUI now runs in Comfy Cloud.
Package your team’s nodes, models and LoRAs (private packs too) into one Build. Every artist opens the same setup in the browser, with nothing to install.
What that means for production teams:
→ New artists start on the same setup as everyone else on day one
→ Roll out a new LoRA or update once, and the whole team gets it
→ Heavier workflows run on RTX PRO 6000, H100, H200 or the 180GB B200, with no hardware to buy
→ Concurrency scales up when clients need it and back down when the work is done
→ Developers run the same setup through the API
Learn more and try your custom Comfy Cloud today with the link below ⬇️
29
LTX VFX capabilities are live on the API and @griptapeai just shared how to get the most out of these capabilities in their platform.
Alpha Gen. SDR to HDR. Reframe. Upscale. Now ready to plug into your team’s pipeline.
Check out their blog: foundry.com/insights/product…
#VFXArtist #AIVFX
W. S. Waldo retweeted
SolidCraft 0.40 is out 🪩
solidcraft.io
We swapped the entire geometry kernel. Every boolean, fillet, shell, sweep and loft now runs on solidcore, our own B-rep kernel in safe Rust.
- Fillets that used to fail now build: variable rounds, three rounds meeting in a corner, Mitre and Swept corners
- New mesher: ~15× faster, and a box is 12 triangles instead of 124
- A plate with 50,000 holes meshes in under 9 seconds
- STEP: one reader, more files open, and a bad face no longer costs you the part
- Cancel actually stops the kernel mid-operation
- Assemblies: place parts by Mate, Flush or Distance
- Notes pinned to the model, sweep along a body's edge, GLB export
- Much more...
W. S. Waldo retweeted
It's been a month since we got the audit notice from Autodesk.
Mutiny BIM is now open source, and it's a pretty cool early release.
Grok Bot wrote a script that built the sheets, schedules, rooms, spaces, RTU, VAV boxes, electrical equipment, circuits and single line diagram, and then it ran the script....and like magic, it all just appeared.
It's very cool, and it's free.
github.com/permitzip/mutinyb…
#Mutiny
It’s here: LTX VFX capabilities are live on the API.
The open-weight VFX workflows you’ve been running in ComfyUI are now production-ready endpoints, with Reframe joining the lineup. Four jobs your pipeline does every week, now one integration.
→ Alpha Gen: a soft, detailed matte from existing footage, no green screen
→ SDR to HDR: meet an HDR delivery spec with the SDR footage you have
→ Reframe: one master, every platform, the scene extended instead of cropped
→ Upscale: lift footage to 4K or 8K, keeping aspect ratio, audio and frame rate unchanged
Available in the LTX API, and through @fal, @ComfyUI and @griptapeai
Try it here: console.ltx.io
Replying to @ltx_io
@ltx_io VFX nodes are now in ComfyUI via Partner Nodes.
→ Reframe to 1:1, 4:5, 5:4, 9:16 or 16:9, with the new area generated to match
→ Upscale to 4K or 8K, keeping frame rate and audio
→ SDR to 10-bit HDR (BT.2020, HLG)
→ Alpha Gen: per-frame mattes up to 4K, with hair and soft edges intact
No local GPU needed 👇
This week, LTX-2.5 dove from orbit, animated a 3D rig, and turned one photo into a film scene. Here are four builds. 🧵
Pick a landmark. Add a character. @jamie_jk’s Earthfall dives from low orbit through the clouds to your character on location.
nitter.gq/jamie_jk/status/211610…
Earthfall: pick a landmark, add a figure, hit Create. You get a dive from low orbit, down through the clouds, to your character at the location.
Runs on your own GPU (API coming soon) with 100% open weights: FLUX 3 Klein, LTX 2.5 and Qwen3.5 on @ComfyUI.
(Compare with Higgsfield Earth Zoom.)
github.com/jamiejk/touchdown
#OpenSourceHiggsfield
W. S. Waldo retweeted
The Mutiny BIM repo is public. Apache 2.0.
github.com/permitzip/mutinyb…
What's in it right now:
- the kernel (elements, types, parameters, units, networks, spaces, phases, sheets, commits)
- trade add-ons for arch, structural, civil, site, mech, elec, plumbing/gas, fire protection, and a pro forma
- plan, 3D and sheet views, routing and sizing, schedules, branches and versions
- IFC 4.3 export and PDF sheets
an MCP tool server so Claude, ChatGPT, Grok or Cursor can run scripts against the model
- live multi-user editing, tested at 250k elements
The kernel doesn't know what a duct or a beam is. Every trade is an add-on on the same contract, so you can write your own.
It's early. It's the engine, not the full product yet. npm install then npm run demo and you're editing a model in the browser.
Go break it. #MutinyBIM
This works extremely well in LMStudio. When coupled with the kosmix/web-tools for web-search, it can turn your local machine into a research engine similar to ChatGPT.
Introducing Arete-27B: Qwen3.8-27B, post-trained to reinforce Western values.
Chinese open models are currently the strongest available, but they ship with state-aligned behavior in the weights. We built an 803-prompt eval across seven value categories, including freedom of expression, democracy, and privacy, and measured how often a model adopts China’s party line.
After fine-tuning on roughly 53 million tokens, Arete-27B's propaganda rate fell from 22% to 1%, in line with the Western models we tested. Its Hermes Bench score shows no significant performance tradeoffs.
The values in an open model are one fine-tune away.
Our eval, weights, and training data are available now along with the full blog post: nousresearch.com/arete
43
W. S. Waldo retweeted
AI dialogue in H3 and Seedance sounds flat. LTX audio fixes it.
Fix Lifeless AI Dialogue: LTX 2.5 Audio for MiniMax H3 & Seedance
I compare native H3 (FastH3) and Seedance 2 dialogue against LTX 2.5 audio
W. S. Waldo retweeted
Introducing Arete-27B: Qwen3.8-27B, post-trained to reinforce Western values.
Chinese open models are currently the strongest available, but they ship with state-aligned behavior in the weights. We built an 803-prompt eval across seven value categories, including freedom of expression, democracy, and privacy, and measured how often a model adopts China’s party line.
After fine-tuning on roughly 53 million tokens, Arete-27B's propaganda rate fell from 22% to 1%, in line with the Western models we tested. Its Hermes Bench score shows no significant performance tradeoffs.
The values in an open model are one fine-tune away.
Our eval, weights, and training data are available now along with the full blog post: nousresearch.com/arete
W. S. Waldo retweeted
mattiaslind.gumroad.com/l/mp…🔶Introducing mPainter(Alpha)!
An open source texturing application with blender at its core, plus a bunch of custom goodies like a new brush-engine etc.
Work with layers, and import the blender materials you already have in a familiar texturing interface😊#b3d #blender #texture #app
W. S. Waldo retweeted
🎮 A new free Blender add-on can turn dense 3D scans and high-poly models into lightweight quad meshes, then automatically check how much of the original shape was lost.
Meet CurioMesh, an open-source retopology tool designed for game artists, sculptors and technical artists.
And unlike traditional one-click remeshing, it lets you redo individual parts without processing the entire model again.
What can it do?
▪️ Automatic quad retopology: Convert dense geometry into lower-poly meshes with mostly quad-based topology.
▪️ Part-by-part processing: Separate a model into meaningful regions and optimize each individually.
▪️ Visual verification: Compare the result with the original from six camera directions.
▪️ Selective corrections: Identify problematic regions and regenerate only those parts.
▪️ Detail baking: Transfer surface information into normal maps and color textures.
▪️ Automatic LODs: Generate multiple levels of detail for game-engine workflows.
▪️ Animation preparation: Build a low-poly cage with higher-resolution detail stored in Blender's Multires modifier.
The workflow is surprisingly practical.
1. Choose your target
Select a preset:
▪️ Low Poly
▪️ Game Asset
▪️ Animation
▪️ Subdivision
▪️ Scan Cleanup
▪️ Sculpt Base
Each preset balances polygon density, surface accuracy and topology quality differently.
2. Reconstruct the topology
CurioMesh divides the geometry into segments, chooses suitable remeshing methods and rebuilds each region.
The user can choose between three layout strategies: Keep Detail, Balanced and Clean Rows.
3. Measure the result
Instead of simply assuming the new mesh is good, the add-on renders comparisons from six viewpoints.
It estimates how much of the visible silhouette and surface appearance changed.
4. Fix only what failed
Review Segments highlights individual regions.
You can change the method, density or geometry constraints for one part without rebuilding everything.
The developer's examples show substantial reductions.
▪️ Concrete road barrier: 60,928 → 987 faces.
▪️ Treasure chest: 103,330 → 10,423 quads.
▪️ Marble bust: 17,456 → 2,508 quads.
▪️ Chess set: 76,832 → 10,110 quads.
The measured visual difference in these examples is around 4–6%, depending on the object.
These are self-reported results, not independent quality benchmarks. The visual-difference percentage also doesn't establish whether the topology is suitable for animation.
And there are more advanced controls.
▪️ Draw guide curves that influence edge flow.
▪️ Define loops around eyes, mouths and joints.
▪️ Paint regions that need more geometric detail.
▪️ Mirror symmetrical topology.
▪️ Apply local Boolean-style changes.
▪️ Preserve UV seams or bake new textures.
Availability:
CurioMesh supports Blender 3.6 through 5.2 on supported platforms.
The add-on runs locally, doesn't upload your geometry and is released under GPL-3.0.
There's also an optional Rust-accelerated processing core in the GitHub release builds.
Important limitations:
It doesn't automatically understand ideal facial deformation topology.
Extreme polygon reduction can damage small details, and smooth surfaces can still produce irregular quads.
Large photogrammetry scans may require minutes or tens of minutes.
Why this matters:
Generating 3D objects is getting easier.
Preparing those objects for actual production is still one of the biggest bottlenecks.
CurioMesh tackles that problem by combining automatic remeshing with measurable visual checks and targeted manual correction.
Making a 3D model is one thing. Making its geometry practical to use is another.
💻 github.com/CurioCrafter/Curi…
#Blender #Retopology #GameDev #OpenSource
W. S. Waldo retweeted
AI video loves to smear the small faces in a scene. this one fixes them by listening to the audio, so lips and teeth match what's said
TaoMate-FVSR can also redub: swap in new speech and the mouth follows. code and weights just went up
taoliveaigc.github.io/TaoMat…
W. S. Waldo retweeted
🎮 AI-generated 3D models often have hundreds of thousands of unnecessary polygons. A new ComfyUI node can rebuild them into clean low-poly meshes with just a few thousand.
Meet ComfyUI-LATO2, a new open-source node that brings the LATO.2 neural remeshing model directly into ComfyUI.
It works with dense meshes generated by TRELLIS.2, Hunyuan3D and other 3D tools.
And unlike traditional polygon reduction, it doesn't simply delete triangles. It rebuilds the geometry from scratch.
Why does this matter?
AI 3D generators frequently produce meshes containing 500K–1M triangles.
They can look impressive, but using them directly in games or real-time applications creates problems:
▪️ Excessive geometry and rendering overhead.
▪️ Poorly distributed polygons.
▪️ Unnecessary internal surfaces.
▪️ Topology that doesn't follow the object's shape.
▪️ Difficult manual cleanup.
ComfyUI-LATO2 addresses these issues using a neural topology-generation model.
How does LATO.2 work?
The original model was developed by researchers from Huazhong University of Science and Technology, Meshy AI and Technical University of Munich.
It separates mesh generation into two stages:
1. V-Flow: Vertex Generation
A learned flow model creates new vertex positions based on the input object's shape and a target vertex count.
You can request anywhere from 200 to 5,000 vertices.
2. T-Flow: Topology Generation
A second model predicts which vertices should connect, producing edges that better follow the object's geometry.
This is fundamentally different from a traditional decimator that simplifies an existing triangle structure.
But the ComfyUI integration goes further.
The original LATO.2 can produce overlapping faces, inconsistent winding and holes.
Developer sakalond added several geometry-processing steps:
▪️ Surface reconstruction: Rebuilds faces from predicted connections, preventing edges from belonging to more than two faces.
▪️ Hole filling: Uses minimum-area triangulation to close gaps.
▪️ Outward orientation: Determines which direction each face should point.
▪️ Internal surface removal: Uses voxel flood fill to eliminate hidden inner walls frequently found in TRELLIS.2 meshes.
▪️ Transform preservation: Restores the original position and scale, allowing the new mesh to replace the source object.
The result?
You can generate a dense 3D object, process it through LATO.2 and export a much lighter GLB without manually rebuilding its topology.
The node integrates with ComfyUI's native MESH type, Preview 3D and Save 3D Model.
Four example workflows are included.
Hardware requirements:
▪️ NVIDIA GPU with at least 8GB VRAM.
▪️ Windows or Linux.
▪️ Approximately 13GB of disk space.
▪️ Python and CUDA-compatible PyTorch.
The installer automatically creates an isolated environment and downloads the required checkpoints.
No manual compilation of CUDA extensions is needed.
Performance:
On RTX 3090/4080 GPUs:
▪️ 30–60 seconds for neural remeshing.
▪️ An additional 1–3 minutes may be needed to remove internal geometry from extremely dense meshes.
Important limitations:
▪️ Geometry only. No UVs, textures or materials are transferred.
▪️ Thin objects can produce folds or self-intersections.
▪️ Narrow openings may get unintentionally closed.
▪️ Different seeds can generate different topology.
▪️ The resulting geometry isn't guaranteed to be suitable for 3D printing or Boolean operations.
The node is released under GPL-3.0.
Why this matters:
AI can already generate impressive 3D objects from images.
But generating geometry and producing optimized assets for a real game engine are two different problems.
ComfyUI-LATO2 helps bridge that gap.
Imagine a workflow where AI creates a model, another model rebuilds its topology, and the result moves into Blender, Godot or Unreal Engine for further processing.
AI-generated 3D is moving beyond making objects look good toward making them practical to use.
💻 github.com/sakalond/ComfyUI-…
🔬 Original LATO.2 research:
lohhhha.github.io/LATO.2/
#ComfyUI #3DGeneration #GameDev #OpenSource























