Nodes/Easycontrol KSampler Compatible/Anima EasyControl (KSampler)
ComfyUI Node

Anima EasyControl (KSampler)

Anima EasyControl image conditioning, KSampler-compatible. Feeds a reference image (VAE-encoded) into a frozen-DiT extended self-attention stream and returns a patched MODEL plus an empty LATENT sized to the conditioning resolution — wire the latent straight into the KSampler so it generates at the (aspect-matched, ~1MP) cond grid. Ideal for spatially aligned tasks like colorization.

By sorryhyun·Created 2 months ago·Updated 24 days ago· 3
Anima EasyControl (KSampler)
  • model
  • vae
  • image
  • mask
  • model
  • latent
base_modelanima
easycontrol_lora
strength1.00
target_megapixels1.0
cond_scale_override0.00
Categoryloaders

Inputs (9)

NameTypeDefaultDescription
modelMODEL
vaeVAE
imageIMAGE
base_modelCOMBOanimaBase model family this node supports. Currently Anima only — the EasyControl reimplementation targets the Anima/Cosmos DiT backbone. More families may be added later.
easycontrol_loraCOMBOEasyControl checkpoint (networks.methods.easycontrol — per-block cond LoRA + b_cond gate). Lives in ComfyUI/models/loras/.
strengthFLOAT1.000–3Conditioning strength. Scales the cond-LoRA deltas (multiplies the trained cond_scale). 1.0 = as trained; raise to push harder toward the reference, lower to loosen.
maskoptMASKOptional inpaint mask. White (selected) region is the hole to regenerate: the node fills it with mid-gray and encodes that as the conditioning image, matching the `inpaint` adapter's training (cond = original image with a gray hole). Wire the ORIGINAL image into the IMAGE socket and the mask here. Leave unconnected for adapters that take a pre-built cond image (e.g. colorize).
target_megapixelsoptFLOAT1.00–4Output size, in megapixels, for the returned LATENT (and the conditioning encode). The image is resized aspect-preserving to ~this many MP and snapped to the VAE/patch grid; the cond stream is encoded at the same resolution so it shares the target's grid (best spatial alignment for colorize). 1.0 keeps generation on the ~1MP distribution Anima was trained at. 0 = keep the input's native resolution.
cond_scale_overrideoptFLOAT0.000–4If > 0, replaces the checkpoint's trained cond_scale (before strength). 0 = use the trained value.

Outputs (2)

NameTypeDescription
modelMODEL
latentLATENT