【FLUX】KONTEXT — v2.0

Use this ComfyUI workflow to edit images with FLUX Kontext: provide a written prompt and an input image, then generate an edited image with optional tiled upscaling.

Load the model, write a prompt, and generate an image; the workflow also supports image-to-image generation.

The workflow contains eight nodes and a built-in SeedVR2 tiled upscaler.

Provide a written prompt and an input image for image-to-image editing.

Saved images include all generation parameters as embedded metadata.

In ComfyUI, turn on “Use Nodes V2 (Vue)” in Settings, refresh the page, and then load the workflow.

Suggestion · non vérifié

The auto version can download models automatically.

Install ComfyUI-UmeAiRT-Toolkit; it is the only custom node to install and packages the upscaler, face detailer, and metadata saver.

For manual base setup, prepare `flux1-kontext-dev-fp8-e4m3fn.safetensors`, `t5xxl_fp8_e4m3fn_scaled.safetensors`, and `clip_l.safetensors`.

For manual GGUF setup, the listed choices are models `G8`, `G5`, or `G4`; CLIP `Q8`, `Q6`, `Q5`, `Q4`, or `Q3`; text encoder `ViT-L-14-TEXT-detail-improved-hiT-GmP-TE-only-HF.safetensors`; and VAE `ae.safetensors`.

Toggle the built-in SeedVR2 upscaler on or off for tiled upscaling.

Use three LoRA slots with individual on/off toggles and strength controls.

Before use, verify that `comfyui-umeairt-toolkit`, `FLUX/Kontext`, and `seedvr2_ema_7b_sharp_fp8_e4m3fn_mixed_block35_fp16.safetensors` are available.

Suggestion · non vérifié

Without ComfyUI’s Nodes 2.0 (Vue) interface, the workflow may have display problems.