日本語版ドキュメントは後半にあります。
This is a UI for inference of ControlNet-LLLite.
ControlNet-LLLite is an experimental implementation, so there may be some problems.
- Clone this repository to
custom_nodes. - Put ControlNet-LLLite models to
ControlNet-LLLite-ComfyUI/models. You can download sample models from here.
Load sample workflow.
You can specify the strength of the effect with strength. 1.0 is default, 0.0 is no effect.
You can apply only to some diffusion steps with steps, start_percent, and end_percent. Specify the number of steps specified in the sampler in steps, and specify the start and end steps from 0 to 100 in start_percent and end_percent, respectively.
(Because we cannot check the total number of steps in the node, this specification has been made. Please check the console output for the specific application range.)
-
If the generated image size is different from the control image size, resize it with
image/upscaling/UpscaleImageas shown in the workflow. -
You can create a Canny image from a normal image with the
image/preprocessors/Cannynode.
This repository is based on IPAdapter-ComfyUI by laksjdjf. Thanks to laksjdjf.
ControlNet-LLLite の推論用のUIです。
ControlNet-LLLiteがそもそもきわめて実験的な実装のため、問題がいろいろあるかもしれません。
custom_nodesにcloneします。ControlNet-LLLite-ComfyUI/modelsにモデルを入れます。サンプルはこちらからダウンロードできます。
サンプルのワークフローを読み込んでください。
strengthに効果の強さを指定できます。1.0でデフォルト、0.0で効果なしです。
stepsとstart_percent、end_percentで拡散ステップの一部にだけ効果を適用できます。stepsにsamplerに指定したステップ数を指定し、start_percentとend_percentにそれぞれ開始と終了のステップを0から100で指定します。
(ノード内で全体のステップ数を確認できないためこのような仕様になっています。具体的な適用範囲はコンソール出力を確認してください。)
- 生成画像サイズと制御用画像サイズが異なる場合はワークフローにあるように
image/upscaling/UpscaleImageでリサイズしてください。 - 通常の画像からCanny画像を作るには
image/preprocessors/Cannyのノードが使えます。
laksjdjf 氏の IPAdapter-ComfyUI を参考にしています。laksjdjf 氏に感謝します。
