MOSS text-to-image gets 85% faster in ComfyUI: 2MP in 90 seconds on 6GB VRAM
cgpixel23 · reddit · 2026-10-22
Reddit user cgpixel23 shared a ComfyUI workflow that massively speeds up the MOSS text-to-image model, tested on an RTX 3060 with 6GB VRAM and 16GB RAM:
- Spectrum nodes + CFG 2.5 + 25 steps: 5 min 42 s (10 min without Spectrum);
- Spectrum + 8 steps + CFG 1.0 + LoRA strength 0.6: just 1 min 30 s for a 2MP image — roughly 85% faster than the standard 25-step config without Spectrum.
Two recommended presets: 8 steps / CFG 1.0 for speed, 25 steps / CFG 2.5 for best quality. MOSS handles photorealistic portraits, artistic styles, poses and typography. The author praises its speed, prompt adherence and style versatility; the workflow file and a Civitai write-up are both public, making it easy for low-VRAM users to replicate.
More from Multimodal
- Magnific Ships Blender Plugin: Generate 3D Assets and PBR Materials from Your Viewport — aziz4ai · 2026-10-22
- Bria AI's Fibo-2.0-dev image model trends on Hugging Face — briaai · 2026-10-22
- WorldStream: a streaming 3D world model with hierarchical planning and persistent state — Xiaoxiao Ma · 2026-10-22
- Seagull steals a burger: a 10-second cinematic video-gen prompt breakdown — michaelrabone · 2026-10-22
- Prompt share: French watercolor character sheets with turnarounds, expressions and poses — azed_ai · 2026-10-22
- AI films look fake because no one directs the light: 6 lighting rules for AI filmmaking — DavidmComfort · 2026-10-22