How to Improve Real Person Face Similarity
Drawingomme · reddit · 2026-07-19
The OP is asking how to achieve true facial resemblance when reconstructing a real person's face using multiple reference images instead of a LoRA. They have tried IP-Adapter, PuLID/InfiniteYou, Flux.2 multi-reference pipelines, and up to 6 images with ReferenceLatents, but the results usually look like "a similar person" rather than a recognizable match. The post seeks more effective practical advice, such as the quantity and quality of reference images, CFG settings, reference strength, FaceDetailer, and better node and base model choices.
More from Multimodal
- Reddit users say Krea 2 Turbo regains strong facial expressions with bypass LoRAs — YentaMagenta · 2026-07-21
- AI music demo blends Suno v5.5, Reason Studios and Grok Imagine 1.5 — Kyrannio · 2026-07-21
- AI anime workflow article breaks storytelling into repeatable prompt steps — Aiden_Tech_Ai · 2026-07-21
- Claude is being pitched as a free workflow for viral YouTube Shorts scripts — Aiden_Tech_Ai · 2026-07-21
- A Bittensor game demo claims a two-person team built a playable 3D world in 30 days — markjeffrey · 2026-07-21
- Gemini Omni Flash turns a boat cabin into a cave inside Flow — chrisfirst · 2026-07-21