Upscaling with Ultimate SD Upscale/Tile Controlnet SD Upscale & Anima Models
基于 Ultimate SD Upscale 或 Tile Controlnet 与 Anima 模型的放大工作流
Thanks @L_A_X for the tile Controlnet
whith new node made by Lebensignal u can use cntile with padding and seam fix features of USDU now.
if ur experience issue with `KeyError: 'AnimaLLLiteApply'`:
make sure u have installed: https://github.com/kohya-ss/ComfyUI-Anima-LLLite
and make this change in ur local 'custom_nodes\ComfyUI-UltimateSDUpscale-FLS-LLLite\modules\processing.py/' : https://github.com/shiranpp/ComfyUI-UltimateSDUpscale-FLS-LLLite/commit/815b2c5d0d1ad0b18494dd9b37bff31b2e2ba3d9
Parameters:
- Prompt: Use the original image's prompt(when doing 1.5-3x upscale with low cn strength), or just quility and style prompt(when doing 3x+ upscale with high cn strength).
- Denoising Strength: Recommended range is 0.25 – 0.75.
- Steps: Adjust according to denoising strength, typically 8 – 20 steps.
- Sampler: er_sde is recommended.
- Controlnet strength: 0.4–1.0; Controlnet end at 0.8 - 1.0
- Plz download CN model from here: https://civarchive.com/models/2708551
参数:
- 提示词：使用原图的提示词(1.5-3倍放大且使用低cn强度)或者只保留质量提示词和画风提示词(3倍以上放大且使用高cn强度)。
- 重绘幅度（Denoising Strength）：建议范围 0.25～0.75。
- 步数（Steps）：根据重绘幅度调整，通常为 8～20 步。
- 采样器（Sampler）：推荐使用 er_sde。
- ControlNet 强度（CN strength）：范围 0.4～1.0，结束位置在0.8-1.0.
- CN tile模型在这里下载: https://civarchive.com/models/2708551
Update Log
- 26/04/12: Version 1.1 updated to preview3.
- 26/04/18: Version 1.1 added a Lora loader and adjusted parameters;
- 26/05/14: Version 1.2 introduced the following QoL (Quality of Life) nodes:
- Auto tile sizing
- Comparison preview
- Save with metadata
- 26/06/21: Version 1.2a adjusted baseline parameters for improved default performance.
- 26/06/21: add first controlnet tile upscale experimental workflow
- 26/06/24: moved to Lebensignal's USDU node that support controlnet lllite
Description
moved to USDU node by Lebensignal that supports controlnet lllite
FAQ
Comments (14)
May I ask, at this stage, is there no similar workflow for Tiled Diffusion like sdxl? The USDU speed is a bit slow.
https://github.com/syntax-one/ComfyUI-Tiled-Diffusion-Anima-Fix. and it is still slow since it just tiled the image in latent compare to tiling image in pixel in USDU/image tile node. u also need to mod the node ur self if u want to use lllite.
@CptCharname oh! thanks! I try it.
谢谢大佬的工作流，我想了解一下有没有anima关于深度图控制和姿势控制的工作流呢？
@CptCharname thank you！！！！
请问target_size 这个是填写原图片的边长吗
分块的大小，直接改USDU里面放大倍率就行了。
@CptCharname 我发现有时候在放大最后一块的时候会出问题，会乱生成一些东西，我原图是832x1216 0.6的重绘幅度 cn强度 0.55
@lingmo 不知道你说的最后一块是什么，你要觉得不符合原图就把cn强度改到0.8-0.9
Ultimate SD Upscale (FLS + LLLite Tile Repair) failed
This node threw an error during execution. Check its inputs or try a different configuration.
post error message here or just post it to AI.
为啥我怎么都跑不起来
用远在Ultimate SD Upscale (FLS + LLLite Tile Repair)
这一步报错,说我缺失了什么东西
发下error message