FluxKontextDev-Lora General workflow通用工作流 — v1.0

This workflow is for the Phantasm series: its listed stages take text and a reference image, process them with FluxKontextDev-Lora, and save an image.

It uses the Flux.1 Kontext base model and includes image scaling, reference-latent handling, LoRA loading, sampling, and VAE conversion.

Guidance, latent initialization, UNET loading, and random noise are also represented by listed nodes.

Text is handled by CLIPTextEncode; LoadImage and ReferenceLatent handle the image and reference; SaveImage handles the image output.

VAEEncode and VAEDecode provide image-to-latent and latent-to-image conversion stages.

Open the workflow on Civitai at https://civitai.com/models/1733176; the listed publisher is Xiaolxl.

Listed model files are ae.safetensors, clip_l.safetensors, flux1-kontext-dev.safetensors, t5xxl_fp16.safetensors, and 幻象绘影HXHY亚洲风格写实-RealisticKontextLoRA.safetensors.

Listed node types include CLIPTextEncode, DualCLIPLoader, LoadImage, ReferenceLatent, EmptySD3LatentImage, FluxGuidance, FluxKontextImageScale, and ImageScaleBy.

Other listed node types are BasicGuider, BasicScheduler, KSamplerSelect, RandomNoise, SamplerCustomAdvanced, ModelSamplingFlux, UNETLoader, LoraLoaderModelOnly, VAELoader, VAEEncode, VAEDecode, and SaveImage.

Image scaling is represented by FluxKontextImageScale and ImageScaleBy.

For the listed LoRA file, the workflow includes LoraLoaderModelOnly.

Before using it, verify that all five listed model files are present in your setup.

Suggestion · not verified

Verify that your setup provides the listed node types, including CLIPTextEncode, DualCLIPLoader, LoadImage, ReferenceLatent, and SaveImage.

Suggestion · not verified