<div data-youtube-video="https://youtu.be/P2bt83Op9bY"><iframe src="https://www.youtube.com/embed/P2bt83Op9bY" width="853" height="480" allowfullscreen></iframe></div> 
 
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, no classifier-free guidance), 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: https://www.runninghub.ai/?inviteCode=rh-v1111

FLUX 3 Klein Base vs Distilled

<div data-youtube-video="https://youtu.be/P2bt83Op9bY"></div> 
 
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, no classifier-free guidance), 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: https://www.runninghub.ai/?inviteCode=rh-v1111

Files with SHA256: 5d0a3e968f88e809ed57ec3f66adf1b2948c328818e6c787e6e5b5c1bf0772e2