FLUX.3 Klein shines at consistent high-res photo editing on RTX 5090: 3000×4000 in ~230s
herbertseabra · reddit · 2026-10-16
A photographer's hands-on test of FLUX.3 Klein: underwhelming for text-to-image generation, but "almost perfect" for consistent photo editing that preserves pose, camera angle, skin texture, and fine details—something FLUX.2 Klein and Qwen 2.1 struggled with at high resolutions. The author calls it genuinely usable for professional photography workflows.
Setup: RTX 5090, 64GB RAM, flux3imagekleinsdbf16.safetensors diffusion model with qwen3vl4bbf16.safetensors text encoder, 16 steps, 230 seconds per image at 3000×4000.
The main downside is speed. All results were achieved without any LoRAs—the author expects dedicated consistency/upscale/sharpen LoRAs could push it further. Comparison image shows high-res skin texture detail.
More from Multimodal
- Claude Opus 5.5 makes a pure-code Halloween animation starring bots trick-or-treating for tokens — sattvi · 2026-10-16
- 1,290-line commit lands MOSS Image support in ostris/ai-toolkit — ostrisai · 2026-10-16
- Ostris AI Toolkit adds training support for OpenMOSS's MOSS-Image-SFT model — ostrisai · 2026-10-16
- Nokia vs iPhone, car edition: AI image shows Tesla Roadster hovering over a Porsche — vaibhavbetter · 2026-10-16
- Midjourney multi-sref weight mixing recipe for cinematic light & motion portraits — michaelrabone · 2026-10-16
- MiniCPM-V 4.7 35B-A3B hits 17+ tokens/s on iPad with ~3GB memory — techNmak · 2026-10-16