Mystic WF - Full flux workflow — v2.0

Mystic WF - Full flux workflow — v2.0 is for creating Flux.1 D images from a text prompt and editing loaded images; it can run img2img and Redux inpainting, use ControlNet Canny and Depth, generate prompts from images with Florence, upscale images, clean Flux LoRA lines, and add

Select ControlNet Depth or Canny, Redux inpainting, img2img, or Florence prompt generation from an image, and bypass sections you do not want to run.

The image-processing path includes first and extra upscaling, LoRA-line cleanup, sharpness and color adjustments, LUT application, and signature or watermark addition.

Use a text prompt, a loaded image, and LoRAs as inputs.

A loaded image can feed img2img, Redux inpainting, Florence prompt generation, or a save step for Civitai information.

Outputs include generated or edited images, upscaled images, saved image files, and Civitai metadata.

Place LUT files in ComfyUI/models/luts, creating the folder if needed, and place upscale models in ComfyUI/models/upscale_models.

Suggestion · non vérifié

The workflow lists these model files: 002_lightweightSR_DIV2K_s64w8_SwinIR-S_x2.pth, 4x_NMKD-Superscale-SP_178000_G.pth, depth_anything_v2_vitl.pth, Flux/ae.safetensors, Flux/clip_l.safetensors, Flux/flux1-dev.safetensors, Flux/flux1-fill-dev.safetensors, and Flux/t5xxl_fp16.s ​​

Named ComfyUI nodes include: ImageBlur, CLIPTextEncode, Bookmark (rgthree), DualCLIPLoader, DifferentialDiffusion, BasicScheduler, Anything Everywhere, ModelSamplingFlux, Canny, InpaintModelConditioning, FluxGuidance, Fast Groups Bypasser (rgthree), LatentUpscaleBy, Image Filter

Additional named ComfyUI nodes are: mxSlider2D, ProPostApplyLUT, Save Images // RvTools, UpscaleModelLoader, ImageResizeKJ, SamplerCustomAdvanced, SplitSigmas, UNETLoader, easy int, GetImageSize+, Any Switch (rgthree), ImageSmartSharpen+, ImageBlend, LoRA Stack to String

Use bypass controls to run one section while finding a seed, then enable additional sections.

Suggestion · non vérifié

Choose which save steps are active and what metadata is saved. The description suggests 0.52–0.62 denoise for the first upscale and 0.1–0.25 for cleanup, with adjustments by style.

Suggestion · non vérifié

Before using it, verify that the listed model files and ComfyUI nodes are available in your setup; if RV Tools is missing, check https://github.com/whitmell/ComfyUI-RvTools.

Suggestion · non vérifié

Adding many LoRAs may produce lines in Flux images; if a LoRA name appears as [object Object], check https://github.com/alcaitiff/ComfyUI-Lora-Auto-Trigger-Words.

Suggestion · non vérifié