【FLUX】IMG to IMG — v2.0

Use this ComfyUI workflow to turn a source image into a new FLUX image with a prompt and Flux.1 D model; it also offers optional SeedVR2 tiled upscaling.

Built with the UmeAiRT Toolkit for ComfyUI, the workflow has eight nodes.

Enter a prompt and use a source image; saved images include generation parameters as embedded metadata for publishing or remixing.

Use ComfyUI’s Nodes 2.0 (Vue) interface: turn on “Use Nodes V2 (Vue)” in Settings, refresh, then load the workflow. It may have display problems if this setting is off.

Install ComfyUI-UmeAiRT-Toolkit through ComfyUI Manager by searching “UmeAiRT,” or use the UmeAiRT Auto-Installer.

Listed requirements include the Image Size to Number and JWDatetimeString node packs, plus ae.safetensors, flux1-dev-Q8_0.gguf, and 4x_NMKD-Siax_200k.pth model files.

For manual base setup, use flux1-dev-fp8.safetensors or flux1-dev-fp16.safetensors with clip_l.safetensors. GGUF setup lists Q8, Q6, Q5, Q4, or Q3 model and CLIP files.

Manual setup also lists t5xxl_fp8_e4m3fn.safetensors or ViT-L-14-TEXT-detail-improved-hiT-GmP-TE-only-HF.safetensors as text encoders, and ae.safetensors as the VAE.

Toggle SeedVR2 on or off to choose whether to use tiled upscaling.

Three LoRA slots have individual on/off toggles and strength controls; you can connect additional LoRA modules.

The auto version is described as downloading models automatically; manual setup offers base and GGUF options.

Before downloading, check that https://civitai.com/models/706253?modelVersionId=880042 shows “【FLUX】IMG to IMG — v2.0” and version “v2.0”.

Vorschlag · nicht geprüft