P-H3-Image draws a 2912x2912 image (8.5 MP) in 15.7 s on my RTX 5090. MOSS needs 8.6 s…
This is a AI post classified by AGK Intelligence as AI image (a model release), kept by the AI Radar because it carries real work, not commentary.
P-H3-Image draws a 2912x2912 image (8.5 MP) in 15.7 s on my RTX 5090. MOSS needs 8.6 s for 1 MP, Qwen 5.0 s. Four image setups, same six prompts, seed 42. The video starts with a zoom into the P-H3 file, then shows every prompt side by side. RTX 5090 (32 GB), median of 6 images, wall clock per image: P-H3-Image w4a8: 15.7 s P-H3-Image int8: 17.9 s MOSS-Image-SFT: 8.6 s Qwen Image 2.1: 5.0 s Size: P-H3 2912x2912 (8.5 MP), MOSS and Qwen 1024x1024 (1.0 MP) P-H3-Image: ComfyUI 0.44.0, Pruna's workflow, 8 steps, own upsampler (3.0 s of the 15.7 s) MOSS: Comfy-Org template, 25 steps, own upsample
Posted by Niklas (2.8k followers) 7 h ago · 3 likes · 305 views · view the original post on X. Kept by the AI Radar as AI image.
More AI work like this
- GPT Image 2.6 is now available in Runway. — @runwayml
- Four directions you didn't think to ask for. — @recraftai
- GPT Images 2.6 is live on Picsart from day one 🎉 Choose Flare for fast iteration or… — @Picsart
- I’ve been loving using Seedance 2.5's Draft Mode in Venice Studio for my recent projects. — @heavypulp
- GPT Image 2.6 — @aimikoda
- We tested Pangram: "Unsure", yet the heatmap found the AI half. Small demo, not a… — @LearnOpenCV
- Opus 5.5 制作 — @ZeroZ_JQ
- The machine with the most memory made my AI picture slowest. 7.3 s on a DGX Spark, 1.7 s… — @niklaslenz_ai
Every post is read and classified by AGK Intelligence: what it is, which market it belongs to, and whether the link is a real tool. 65.6k posts from 5k X accounts over the last 14 days, 10.4k tools, 19 markets. New posts are collected several times an hour and the full ranking is recomputed every hour — last update 2026-11-06 16:05 UTC. Full method.