Flux.2 Klein 9B identity drift: seeking fixes for cinematic character consistency
RecordEmbarrassed787 · reddit · 2026-09-07
A creator building a cinematic documentary pipeline with Flux.2 Klein 9B (ComfyUI + fal.ai API on an RTX 5090 Laptop) reports that character identity collapses whenever prompts shift to complex combat scenes or dramatic lighting — the native ReferenceLatent conditioning can't anchor facial features.
Current pipeline: reference image → upscale to 1MP → VAE Encode → ReferenceLatent conditioning → sampling via fal.ai → decode. No custom LoRA is used yet.
Options being weighed: masking the face before VAE Encode, trying ReferenceLatentPlus, or dropping fal.ai entirely and running a local Flux-PuLID/FaceDetailer pass to restore identity after API output.
Related event: Flux.2 Klein 9B Faces Identity Drift in Cinematic Pipelines(2 posts)→
More from Multimodal
- Sheet Music to MIDI Pipeline Offers a More Structured Space for Music Models Than Audio — felpix_ · 2026-09-07
- Seen2Scene open-sourced: flow matching trained on real partial scans for 3D scene completion, ECCV 2026 — angelaqdai · 2026-09-07
- TAIKEN: a 4-minute AI action film made with MiniMax H3 and ChatGPT — bennash · 2026-09-07
- Midjourney watercolour-style generations showcased — davidhoang · 2026-09-07
- Wuxia "Netherworld Divine Palm" AI video made with open-source MiniMax H3, prompt shared — Low_Loquat_4035 · 2026-09-07
- ComfyUI-Easy-Use adds Pass or None node for conditional value routing in workflows — altoiddealer · 2026-09-07