Flux.2 Klein Image Edit Workflow — V2.0

Flux.2 Klein Image Edit Workflow — V2.0 is a ComfyUI workflow for editing a loaded image with Flux.2 Klein 4B; it takes an image, text, a reference latent, a seed, and random noise, and can output an edited image that can be upscaled, compared, and saved.

The workflow encodes and decodes images with a VAE, then samples with text conditioning, a reference latent, a seed, and random noise.

Version V2.0 adds an upscaling path and includes image comparison and image saving.

Use LoadImage, CLIPTextEncode, ReferenceLatent, Seed (rgthree), and RandomNoise for the image, text, reference-latent, seed, and noise inputs.

VAEDecode produces the decoded image; ImageUpscaleWithModel, Image Comparer (rgthree), and SaveImage provide upscaling, comparison, and saving stages.

The listed custom nodes are ComfyUI-Manager, rgthree-comfy, and ComfyUI-GGUF.

Place the diffusion model under .../ComfyUI/models/diffusion_models/, the text encoder under .../ComfyUI/models/text_encoders/, and the VAE under .../ComfyUI/models/vae/.

Suggestion · not verified

The workflow lists these required model files: 4x_foolhardy_Remacri.pth (upscaler), flux-2-klein-4b-BF16.gguf and flux-2-klein-4b-fp8.safetensors (UNet), flux2-vae.safetensors (VAE), and qwen_3_4b.safetensors (text encoder).

The named node packs are rgthree-comfy and comfyui-gguf.

For the distilled setup, use 4 steps, CFG 1.0, and the Euler sampler.

Suggestion · not verified

The listed model choices include flux2-klein-4b or other diffusion-model versions, qwen_3__4b or qwen_3__8b, and flux2-vae.

Before setup, compare the listed choices—flux2-klein-4b, qwen_3__4b, and flux2-vae—with the exact files you install: flux-2-klein-4b-BF16.gguf or flux-2-klein-4b-fp8.safetensors, qwen_3_4b.safetensors, and flux2-vae.safetensors.

Suggestion · not verified

Before setup, verify which package provides comfyui-gguf and whether qwen_3_4b.safetensors is the file you intend to use for the qwen_3__4b text-encoder choice.

Suggestion · not verified