FLUX 3 Image Klein layout-controlled editing workflow. Every element gets a box: you say where it comes from in the reference and where it should land in the output.
Main features:
- Built on the distilled FLUX 3 Image Klein checkpoint (flux3_image_klein_sd_int8_convrot + qwen3vl_4b text encoder + flux2 VAE)
- CreateBoundingBoxes defines src_bbox / tgt_bbox pairs on a 0-1000 normalised grid, so the same box works at any output resolution
- Flux3LayoutPrompt converts those boxes into the caption syntax the model expects (<object_1>, <object_2>, ...)
- ReferenceMultiLatent keeps the source image conditioning alive while regions move
- 16 steps, no classifier-free guidance, euler/simple, denoise 1.0
Suggested workflow:
1. Load the reference image and draw a source box on the element you want to move, plus a target box where it should end up.
2. Pin everything you do not want to move by giving it an identical src and tgt box - an unmoved box is a "keep" instruction.
3. Write the caption around the boxes and run. Bounding boxes bias placement strongly but are not a hard mask, so expect small drift.
Dependencies:
- flux3_image_klein_sd_int8_convrot.safetensors (distilled UNET)
- qwen3vl_4b_int8_convrot.safetensors (text encoder / CLIP)
- flux2-vae.safetensors (VAE)
- CreateBoundingBoxes, Flux3LayoutPrompt, ReferenceMultiLatent (custom nodes)
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