FLUX 3 Image Klein official image-editing workflow. Upload a reference image, draw boxes over the regions you want changed, and the model rewrites only those regions while keeping the rest pixel-faithful.
Main features:
- Built on FLUX 3 Image Klein base (flux3_image_klein_base_int8_convrot + qwen3vl_4b text encoder + flux2 VAE)
- Bounding-box editing: CreateBoundingBoxes marks src/tgt regions and Flux3LayoutPrompt references them as <object_1>, <object_2> in the caption
- Base model path: 50 steps, CFG 3.5, euler/simple, denoise 1.0 - the highest-quality setting before distilled shortcuts
- Multi-image reference support through ReferenceMultiLatent and BatchImagesNode: feed up to 10 references and address them as ref_image_0, ref_image_1, ...
- Prompt upsampler (qwen3.5_9b_flux3_prompt_upsampler) is included and can be bypassed when you already have a polished caption
- ComfySwitchNode lets you toggle between edit strengths in the same graph
Suggested workflow:
1. Load the source image. Draw one or more boxes on the parts you want to change.
2. Describe the final picture completely - not just the change. Name the boxes (object 1, object 2) and say what each one should contain.
3. Run the graph. Compare with ImageCompare and iterate. Keep everything outside the boxes unchanged by leaving it out of the prompt.
Dependencies:
- flux3_image_klein_base_int8_convrot.safetensors (base UNET)
- qwen3.5_9b_flux3_prompt_upsampler_w4a8.safetensors (prompt upsampler)
- qwen3vl_4b_int8_convrot.safetensors (text encoder / CLIP)
- flux2-vae.safetensors (VAE)
- BatchImagesNode, ComfySwitchNode, CreateBoundingBoxes, Flux3LayoutPrompt, ReferenceMultiLatent, TextGenerate (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