Blog
Behind-the-scenes logs, 3D workflow breakdowns, and project announcements from the creative front.
Which compact 4B model offers the better local workflow? Across 30 matched configurations on an RTX 3060 12GB, Mage-Flow generated images faster, while FLUX.2 klein produced cleaner results and the better overall quality/speed balance.
A 60-image ComfyUI benchmark comparing generation speed, memory usage, character fidelity, photography and image quality on an RTX 3060 12GB.
A technical resolution benchmark of Krea 2 Turbo scaling from 1MP to 4MP with 12GB VRAM
A detailed timing and quality comparison benchmark of Krea 2 Turbo vs. Raw vs. Raw + Turbo LoRA on an RTX 3060 12GB under ComfyUI.
Rigorous technical benchmark, visual comparisons, and iteration sweet-spots of Krea 2 Turbo in ComfyUI under low-VRAM constraints, optimized for 12GB graphics cards.
A low-VRAM Musubi Tuner workflow for Krea 2 Raw training and Krea 2 Turbo inference.
A practical low-VRAM guide to training Ideogram 4 LoRAs locally with AI Toolkit, FP8 models, JSON captions, and a 12GB GPU.
Ideogram 4 responds extremely well to structured JSON prompts. This guide explains how to build clean JSON prompts for ComfyUI workflows, how to use the Ideogram 4 Prompt Builder KJ node for layout control, and why composition, bounding boxes, and visual evidence matter more than long negative prompt lists.
MediaPixel Games is now open: a new space for 3D assets, game art experiments, tutorials, production notes and downloadable resources.