RefMod tuning on RTX 5090: 20 reference images burn 18,500 tokens for marginal speed gains
spacemidget75 · reddit · 2026-09-23
A Reddit user with an RTX 5090 is seeking help tuning character reference (RefMod) settings for face and full-body consistency: AI assistants recommend 4-6 images at 8,000 tokens, while HF docs allow up to 20.
- Their 20-image + 2 full-body rotation video setup costs 18,500 tokens
- Generation slowed only from 43s/it to 53s/it, unclear if more images help quality past a point
- Still unsure whether to render at 1024 or 2048 and how to set other parameters
No resolution yet, but useful context for anyone doing character-consistent generation.
More from Multimodal
- Dev builds a character-swap LoRA dataset end-to-end with Codex and GPT Image 2.5 — ostrisai · 2026-09-23
- Tencent Hunyuan Image3.5 preview went live Sept 22, free for a limited time — HeyAmit_ · 2026-09-23
- Hands-on with Tencent's Hy Image 3.5: strong text rendering, editing and 5-image references — HeyAmit_ · 2026-09-23
- ImIR replaces text prompts with image instructions for all-in-one restoration — Süleyman Aslan · 2026-09-23
- Alibaba's Qwen Launches Five-Model Audio Stack, Slashes TTS 70% and ASR up to 95% — Alibaba_Qwen · 2026-09-23
- Microsoft ships MAI-Transcribe 2 with multilingual speech, diarization, timestamps — lee_stott · 2026-09-23