Daily versatility-easy get incredible ComfyUI workflow — v2.1(Llama 3.1 Polish up) — v2.1(Llama 3.1 Polish up)

This ComfyUI workflow is for creating and refining images from Chinese or English text prompts and loaded images; it can generate or polish prompts with LLMs, produce detailed reverse-image descriptions with LLaVA, and expose named image stages including base image, base image2,

The text-to-image path includes prompt completion in Chinese and English, ControlNet, IPAdapter, portrait-master prompts, instantID, PAG attention guidance, detail repair, high-definition repair, and a Kolors polishing option.

A separate background-replacement path uses Change background_basis and Change background_repair as named image stages.

Use a Chinese or English prompt for text-to-image, or load an image for reverse image reasoning.

Named image stages are: base image after first generation; base image2 after Detail Repair; HD image after HD Repair; HD image2 after Kolors Zoom in; Change background_basis after Background Replacement; and Change background_repair after Background Replacement: Repair.

Ollama runs separately from ComfyUI; download Ollama, Llama 3.1, and LLaVA for prompt generation or polishing and image reverse description.

Some nodes download models automatically, while others require manual download and placement.

Named node-pack components include: Image Resize; easy preDetailerFix; FreeU_V2; VAELoader; SUPIR_encode; KolorsSampler; ShowText|pysssss; ImageUpscaleWithModel; TomePatchModel; OllamaVision; SUPIR_conditioner; easy promptConcat; DownloadAndLoadChatGLM3; easy instantIDApply; SUPサ

More named node-pack components include: easy showAnything; easy ipadapterApplyADV; SUPIR_first_stage; StringConstantMultiline; easy applyBrushNet; SUPIR_decode; SUPIR_model_loader_v2; easy preSampling; easy stylesSelector; LoadFlorence2Model; KolorsTextEncode; VAEEncodeTiled; In

Named model files and identifiers include: Kwai-Kolors/Kolors; SUPIR-v0Q.ckpt; Dream Tech XL _ 筑梦工业XL_v5.0 - 与光同尘.safetensors; xinsir-controlnet-union-sdxl-1.0ProMax Model.safetensors; sam_vit_b_01ec64.pth; 4xNMKDSuperscale_4xNMKDSuperscale.pt; kolors_vae.safetensors; realDreamSD

Use the easy use management group to enable or ignore node groups as needed, and read each group’s notes before using it.

Suggestion · not verified

For normal text-to-image, turn off background-replacement node groups; when switching paths, select the matching image in the Get node and connect Change background_basis or Change background_repair to the picture input of the pre-detail repair node.

Suggestion · not verified

Opening too many node groups at once can exhaust video memory; enable groups according to your hardware.

Suggestion · not verified

If the full process is too large for your hardware, complete earlier steps first, close groups that are not needed later, and copy images into later links to continue.

Suggestion · not verified