📡 Research Dispatches
Field notes from the computational frontier, covering the practical mechanics of local inference, GPU limits, workflow engineering, and AI systems worth documenting.
103 articles · newest firstIn the undeterministic universe, knowledge is the only asymmetric advantage that compounds. These dispatches are generated from the research pipeline of The Freeloader's Guide — each article a wildcard * probe into the shifting landscape of AI compute, local inference, VRAM optimization, and the tools that let you run the machine on your own terms.
From ComfyUI workflows to GPU architecture deep-dives, from Stable Diffusion optimization to the economics of the AI compute stack — this is the raw signal from the frontline of the Computational Apocalypse. Every piece exists because the wildcard matched something worth documenting.
Start here: practical ComfyUI guides
Choose the problem you want to solve. These guides explain the method, link to sources and state what has and has not been tested.
- Fix slow ComfyUI runs and VRAM failures
- Understand RTX 4090 memory limits and alternatives
- SDXL con poca VRAM: diagnosticar errores y comparar cambios
- Find workflow references and check their evidence status
- Explore the ComfyUI ecosystem source map
Working with video? Check a YouTube video's category with TubePilot (your own YouTube API key is required).
- Engineering Log: ByteDance DreamActor M2.0 Architecture & Pipeline Integration
- Engineering Log: DreamActor M2.0 vs Kling 2.6 Architecture Analysis
- DreamActor-M2: architecture and evidence boundaries
- Mocha Wan Workflow: Character Replacement in ComfyUI
- Engineering Generative Systems: Verified 2026 Field Guide
- Engineering Log: Deploying Generative Microsites via Promptus Architecture
- Hunyuan 1.5 Technical Analysis: Optimizing DiT Architectures on Consumer Hardware
- Scaling Local Inference: Optimizing Flux and SDXL Workflows
- FLUX.2 Klein 4B and 9B: Verified Model and VRAM Facts
- New Lip Sync with AI Avatar is So Good - Check Out Demo
- Promptus AI - Tutorial - How To Make An AI Image
- Run Unlimited AI Generations on Your Own GPU Server
- Sora 2 Video Example — Watch Full Video in Channel
- This New Mocha Wan Model is INSANE (ComfyUI → Promptus Workflow Tutorial)
- Wan 2.1 Install Locally on PC using 8GB VRAM - Promptus
- 4K AI Images That Look REAL (Dype ComfyUI Workflow in Promptus Tutorial)
- Organising a Pony-compatible ComfyUI graph without hidden dependencies
- ComfyUI in Promptus: Install → Generate → SAVE Workflows (Full Tutorial)
- FLUX.2 klein for interiors: evaluate layout fidelity before surface detail
- How to Download Promptus with ComfyUI
- What ComfyUI Is Really Used For
- AI infrastructure decisions: distinguish commercial proposals from contract terms
- Reading AI monetisation news without confusing announcements and rollout
- Local inference versus a hosted API: compare cost per accepted deliverable
- A memory budget for mixed local and hosted AI workflows
- OpenAI’s Commercial Pivot and the 2026 VRAM Optimization Stack
- Migrating an AI workflow without losing its acceptance criteria
- OpenAI’s Commercial Speedrun and the 2026 VRAM Efficiency Protocol
- Portable AI project records: preserve more than the prompt
- OpenAI’s Monetization Pivot & 2026 VRAM Optimization Logic
- Validate a lower-memory model variant before adopting it
- ChatGPT advertising and an AI application's architecture are separate questions
- Instrument an AI workflow's cost before optimising it
- Data boundaries for a local-first AI workflow
- Plan capacity for bursty AI workloads without hiding queue time
- Upgrade a ComfyUI environment with a tested rollback path
- Hosted AI vs Local Inference: An Evidence-Led Method
- Testing a snapshot-style LoRA on a memory-constrained workstation
- Qwen 3 TTS Deployment: Local Audio Generation and Multi-Character
- Qwen3-TTS dialogue: measure first audio and full completion separately
- Mastering Qwen3 TTS in ComfyUI: Advanced Voice Synthesis and Cloning
- An engineering handoff for reviewing AI service obligations
- Optimising Qwen3 TTS on ComfyUI: Low-Resource Deployment Strategies
- Qwen3 TTS in ComfyUI: Advanced Voice Cloning & Control
- Qwen3 TTS in ComfyUI: Advanced Voice Synthesis & VRAM Optimisation
- Before signing up to a free AI image website: an evidence checklist
- Compare free image generators with a task-specific test set
- Evaluate voice design and consented cloning before publishing speech
- Z-Image-Turbo: understand the local workflow before evaluating speed
- Fast Stable Diffusion Install: ComfyUI Power!
- Install Stable Diffusion: A ComfyUI Engineer's Guide
- SDXL Low-VRAM Troubleshooting in ComfyUI
- SDXL for Beginners: ComfyUI Low VRAM Guide
- Reading a Stable Diffusion workflow from input to saved image
- Z-Image-Turbo: High-Speed Image Generation in ComfyUI
- AI's Misunderstood Reality: ComfyUI Deep Dive
- ComfyUI: A Definitive Guide for Advanced Users
- ComfyUI Basics: Nodes, Models, and Workflows
- ComfyUI Deep Dive: Workflows, Nodes & Optimisation
- ComfyUI: Install & Optimize Stable Diffusion
- ComfyUI: Install, Use & Optimize Stable Diffusion
- ComfyUI: Installation, Models & First Image
- ComfyUI: Installation, Workflows & VRAM Tricks
- ComfyUI Intro & Install: SD on a Budget
- ComfyUI: Master AI Image Generation - A Deep Dive
- ComfyUI: Master AI Image Generation Workflows
- ComfyUI Image Generation: Install, ControlNet and Upscaling
- ComfyUI: Your Definitive Install & Workflow Guide
- Install Stable Diffusion: 60-Second Guide
- Low VRAM SDXL: ComfyUI Optimization Tactics
- Mastering ComfyUI: A Deep Dive for AI Image Generation
- Stable Diffusion: A 2026 Beginner's Guide
- Auditing a claimed 48GB RTX 4090 benchmark
- RTX 4090: reduce a workflow's memory demand before changing hardware
- RTX 4090 VRAM: 24GB limits, 48GB claims and practical alternatives
- Evaluating an advertised modified GPU without relying on hype
- Double Your RTX 4090 VRAM: The Mod Scene Deep Dive
- Choosing a ComfyUI compute route when an RTX 5090 is not available
- RTX 5090 in ComfyUI: 32GB VRAM versus 128GB system RAM
- Scaling SDXL on an RTX 5090 with a measured workload envelope
- SDXL "Easy Workflow" in ComfyUI: 2026 Guide
- ComfyUI SDXL Performance: Speed and VRAM Tuning
- ComfyUI Tiled Diffusion Setup for High-Resolution SDXL
- ComfyUI Tiled Diffusion: High-Res Without the VRAM
- Planning an illustrated story with SDXL on a constrained workstation
- SDXL ComfyUI Workflow: Lightning Fast!
- SDXL Easy Workflow: ComfyUI in 2026
- SDXL Easy Workflow: Lightning-Fast ComfyUI in 2026
- SDXL Stylized Image Generation in ComfyUI
- SDXL Tiled Upscale: The Definitive 4K Guide
- SeedVR2 vs SDXL Tiled: 4K Upscaling Deep Dive
- Tiled Diffusion: Fix SDXL VRAM Issues in ComfyUI
- AI Claudia Kiss: Scaling SDXL on Modest Hardware
- Observe ComfyUI memory use before changing the workflow
- ComfyUI RTX performance: fix slow runs and VRAM failures
- Pinokio AI: Advanced Workflows & Optimizations (2026)
- Building a consistent cartoon shot sheet in ComfyUI
- ComfyUI parameter control: a reproducible test method
- A creative decision journal for ComfyUI projects
- Hunyuan3D in ComfyUI: distinguish shape generation from texturing
- Flux2Klein: Real-Time AI Image Editing on a Budget
- Reviewing image enhancement in ComfyUI without confusing detail with accuracy
- SDXL on Budget: VRAM Optimization with ComfyUI & Promptus AI