FLUX 3 Klein Base vs Distilled

FLUX 3 Image Klein A/B workflow: base and distilled checkpoints run side by side from the same seed, so you can judge exactly what the speed-up costs. Main features: - Two checkpoints in one graph: flux3_image_klein_base_int8_convrot (25 steps, CFG 3.5) and flux3_image_klein_sd_int8_convrot (16 steps, CFG 1), sharing one qwen3vl_4b text encoder and one flux2 VAE - Identical latent, identical seed, identical caption - only the model and sampler settings differ, so the comparison is fair - ImageCompare renders both outputs next to the source for direct judging - Demonstrates the practical difference: the base path is slower and slightly more literal, the distilled path is roughly 3x faster with close-but-not-identical detail - No LoRA and no upscale, so you are comparing raw checkpoints Suggested workflow: 1. Write one detailed caption and set a fixed seed. 2. Run the graph. Both checkpoints render; ImageCompare lines them up. 3. Judge by task, not by score: portraits, product text and pose edits usually favour one path consistently, so run a handful of your own subjects before deciding. Dependencies: - flux3_image_klein_base_int8_convrot.safetensors (base UNET) - flux3_image_klein_sd_int8_convrot.safetensors (distilled UNET) - qwen3vl_4b_int8_convrot.safetensors (text encoder / CLIP) - flux2-vae.safetensors (VAE) - ImageCompare (rgthree Node Suite) RunningHub workflow page: https://www.runninghub.ai/zh-cn/user-center/1846601088051884033/webapp?inviteCode=rh-v1111 Bilibili tutorial video: https://www.bilibili.com/video/BV1btJb6LEGS/ RunningHub registration with invite code rh-v1111 (1000 RH coins): https://www.runninghub.ai/?inviteCode=rh-v1111