Hi Everyone!

I wanted to share a technical showcase of a standalone local 3D pipeline project I've been building, called *Standalone Local Orchestration Platform*

I built this to solve a major bottleneck: going from an initial prompt/image concept to a fully rigged, animated 3D asset ready for game engines without relying on cloud subscription services, queue times, or complex terminal setups.

### 🛠️ How the pipeline works:

1. **Text/Image-to-3D Baking:** Generates textured 3D meshes locally using Microsoft's Trellis 2.0 framework.
2. **Automated Base Rigging & Retargeting in Houdini:** The system automatically generates a base Rigging and Retarget setup inside Houdini, giving technical artists full control to manually fine-tune joint placements and weights if needed.
3. **1-Click Unreal Engine 5 Export:** Transfers the fully textured and animated character straight into UE5 via automated local RPC bridges.
4. **Smart Folder Organizer:** Sorts 2D concepts, 3D meshes, prompt metadata, and turntable preview GIFs into clean subdirectories (`PackName/Asset_Timestamp/`) without zip bloat.

### 💡 Under the Hood (Backend Transparency):
The manager is built on top of **ComfyUI** and **Ollama**. Ollama handles the local prompt engineering and concept logic, while ComfyUI drives the underlying image generation and Trellis 3D baking. To prevent Out-of-Memory (OOM) crashes on consumer GPUs, the app automatically flushes local Ollama VRAM (`keep_alive: 0`) before executing ComfyUI generation cycles.

The application is compiled into a native Windows binary using Nuitka for speed and portability, running 100% offline on `127.0.0.1`.

I would love to hear your thoughts on this orchestration approach!

Hello Everyone!

I wanted to share a sneak peek of the upcoming **3D Asset Edition** of my standalone launcher, the **S.L.O.P. Manager** (Standalone Local Orchestration Platform).

To be completely transparent: **this app is built on top of ComfyUI!**

I got tired of terminal environment conflicts, manually managing custom node dependencies for heavy models, and dealing with massive VRAM clashes. So I built a compiled, standalone Windows app (compiled with Nuitka) that automates and orchestrates the backend.

### What runs under the hood & how it works:
* **ComfyUI Portable Backend:** On first boot, the app boots up a local ComfyUI Portable environment. It executes the workflows (currently running the Trellis node package for 3D generation and Qwen-VL for local image editing) entirely via WebSockets/API.

* **The Ollama VRAM Shield:** Running local LLMs (Ollama) and heavy ComfyUI nodes (Trellis) on a single consumer GPU is a VRAM nightmare. The manager automatically intercepts the generation call, flushes the local Ollama memory (`keep_alive: 0` API call), sleeps for 1 second to clear the GPU, and then triggers the ComfyUI API to prevent Out-of-Memory (OOM) crashes.

* **Smart Folder Organizer:** Moves the 2D source, the textured GLB model, prompt metadata, and the auto-rendered turntable GIF from the output directory into tidy structured folders (`PackName/Asset_Timestamp/`) without bloated zip archives.

* **Staging Queue (Local RPC Bridges):** It acts as an external bridge. You can select your generated GLB assets and send them directly into **Unreal Engine 5** or **Houdini** via automated local RPC connections.

All the underlying workflows are standard ComfyUI JSON files stored in the workspace directory. I’m focusing on making this a clean "1-click" experience for artists who love ComfyUI's power but want to bypass the node-spaghetti during high-volume asset production.

I'd love to hear your feedback on this orchestration approach!

Hello Everyone 👋

Setting up ComfyUI workflows, managing 30GBs of .safetensors files, and writing perfect prompts is a nightmare for non-technical users. So I tried to make a bit more user friendly UI around ComfyUI

I wanted to build something that my non-coder friends could use to generate high-quality AI art locally, without paying for cloud subscriptions. So, I built the AI S.L.O.P. Manager (Standalone Local Orchestration Platform).

It’s a completely local GUI built entirely in Python.

🔗 GitHub Repo: https://github.com/Tamerygo/ai-slop-manager-starterEdition

🛠️ Under the Hood (The Python Stuff)

The whole app is orchestrated using Python, acting as a bridge between Gradio 6.0 (Frontend), Ollama (Local LLM for prompt engineering), and ComfyUI (Image generation backend).

Here are some of the cool Python solutions I implemented:

• Zero-Click Auto-Bootstrap: 
• Feature-Driven Setup: Instead of asking users to download "Juggernaut_XL_v9.safetensors", the UI asks: "Do you want the 📸 Photorealistic Studio capability?". Python calculates the required disk space (shutil.disk_usage), checks existing files, and downloads the exact models via HuggingFace streams directly into the correct ComfyUI folders.
• VRAM Watchdog & Token X-Ray: To prevent 16GB GPUs from crashing, the app has a custom token estimator. If a user's prompt exceeds 250 tokens, Python automatically routes the prompt to a local qwen2.5-coder:3b model to compress and optimize it before sending it to ComfyUI. It also forces a VRAM flush (keep_alive: 0) between batches.

The whole thing is packaged into a portable Windows executable.

The "Starter Edition" is completely free to try

Let me know what you think! 🚀

I couldn't find a single ComfyUI custom node that produces a usable turntable straight from a Trellis mesh — so I built one and finally cleaned it up enough to share.

It renders clean turntables (and Front/Side/Back views) directly from Trellis meshes, all inside ComfyUI. No exporting to external tools, no round-tripping through Blender.

Along the way I also fixed the texture seams that kept showing up: the renderer was antialiasing the raw UV coordinates, so along every UV island edge the interpolated UV pointed to a random spot in the atlas — hence the seams. This node renders the UV pass without antialiasing and cleans up edges with supersampling (SSAA) on the final textured image instead, plus a nearest-sampling fallback right on the seam pixels. Result: no more seams.

Under the hood it's a "stealth" adapter around the Trellis renderer — it skips the PBR path (which tends to hang) and does texturing + lighting in native PyTorch instead.

Features:

• Front / Side / Back / full Turntable modes
• Adjustable camera (distance, elevation, FOV) and simple ambient + directional lighting
• White / Black / Gray / Transparent background
• Exports per-frame camera extrinsics/intrinsics as JSON
• ssaa and seam_fix toggles for quality vs. speed

Note: it's not standalone — it assumes you already have a working Trellis workflow that outputs a mesh. This node just handles the rendering/turntable step on top of it.

GitHub + install instructions: https://github.com/Tamerygo/ComfyUI-TrellisNativeTurntable

Feedback welcome,

Game Title: Viking Striker

Playable Link: https://tamerygo.github.io/viking-striker/

Platform: Browser (HTML5 — plays on desktop & mobile)

Description: Viking Striker is a free one-tap football arcade that runs right in

your browser, on desktop or phone, with no download and no sign-up. You start by

picking your nation from a row of flags, then choose the opponent whose defenders

you'll take on — or use the quick-pick buttons to drop straight into a real matchup.

Two modes share one simple control: in FLY mode you tap to float your striker up

through the gaps between towers of defenders, and in RUSH mode your player runs

automatically while you tap to jump and double-jump over sliding tackles. The gaps

get tighter as your score climbs, some defender towers start moving, and a few

defenders even charge straight at you — so it stays tense the whole way. Beat your

best, hit share, and challenge a friend. It's a single lightweight HTML file with

hand-drawn chibi characters and a bit of stadium atmosphere, made during the World

Cup as a quick, fun side project. I'd love to hear what score you reach and whether

the difficulty feels fair.

Free to Play Status:

[X] Free to play

[ ] Demo/Key available

[ ] Paid (Allowed only on Tuesdays with [TT] in the title)

Involvement: I'm the solo developer — This is my first small game, so honest feedback is genuinely welcome.