ComfyUI-SD3-Powerlab

G370SD3PowerLab_AttentionToImage

Extracts the selected SD3 joint-block backbone QKV attention weight tensor and reshapes it into an image tensor.

SD3 Power Lab/Hack · 1 input · 2 parameters · 1 output

G370SD3PowerLab_ImageIntoAttention

Patches a cloned SD3 model's selected joint-block backbone QKV attention weight using an attention image and the supplied patch and model strengths.

SD3 Power Lab/Hack · 2 inputs · 4 parameters · 1 output

G370SD3PowerLab_ImageIntoLayer

Patches a cloned model tensor selected by a layer substring using an image converted to the chosen tensor dimension.

SD3 Power Lab/Hack · 2 inputs · 4 parameters · 1 output

G370SD3PowerLab_LayerToImage

Renders a selected model-state tensor as an image, handling 1D and 2D tensors and the specified 3D diffusion_model.pos_embed case.

SD3 Power Lab/Hack · 1 input · 1 parameter · 1 output

G370SD3PowerLab_RenderAttention

Renders a selected SD3 joint-block backbone attention weight as query, key, value, all-stacked, or all-interposed output, with a selectable colormap.

SD3 Power Lab/Visualize · 1 input · 4 parameters · 1 output