ComfyUI-STARFlow

STARFlowCheckpointLoader

Loads a STARFlow transformer model checkpoint with configuration

STARFlow · 1 output

STARFlowSampler

Generates latent samples using STARFlow reverse flow

STARFlow · 4 inputs · 14 parameters · 1 output

STARFlowT5TextEncode

Encodes text using T5-XL (google/flan-t5-xl) encoder with STARFlow-specific metadata

STARFlow · 1 input · 5 parameters · 1 output

STARFlowVAEDecode

Decodes STARFlow latents to images

STARFlow · 2 inputs · 1 parameter · 1 output

STARFlowVCheckpointLoader

Loads a selected STARFlow-V checkpoint using the selected dtype and returns a STARFlow-V model wrapper.

STARFlow V · 1 output

STARFlowVDecode

Thin convenience wrapper around a standard core VAE's decode(), provided only so a STARFlow-V workflow can be expressed with suite-native node names. Byte-for-byte equivalent to feeding STARFlowVSampler's LATENT output directly into the core VAEDecode node with a normally-loaded Wan2.2-TI2V VAE -- both consume the same raw (un-bridged) standard LATENT layout and apply the same batch-reshape as comfy's own nodes.py::VAEDecode.

STARFlow V · 2 inputs · 0 parameters · 1 output

STARFlowVImageEncode

Encodes a resized, center-cropped image with a Wan2.2-TI2V video VAE and returns a latent for the first frame.

STARFlow V · 2 inputs · 3 parameters · 1 output

STARFlowVSampler

Runs STARFlow-V's block-wise reverse (autoregressive-flow inversion) sampling: deep (fully causal, text-conditioned) blocks first, then shallow (temporally-local) blocks, exactly matching upstream Model.reverse -> reverse_deep -> reverse_shallow. Supports both the default block-wise Jacobi iteration decoder and the (slower) strictly sequential ancestral decoder as a fallback/debugging mode. Output is a *raw*, un-normalized standard ComfyUI video LATENT (B, C, T, H, W) -- i.e. it is bridged back out of STARFlow-V's internal normalized + temporally-scaled latent space -- so it can be fed directly into the standard core VAEDecode node (or this suite's STARFlowVDecode, which is a thin equivalent) using a normally-loaded Wan2.2-TI2V VAE.

STARFlow V · 3 inputs · 13 parameters · 1 output

STARFlowVTextEncode

Encodes a prompt with FLAN-T5-XL using STARFlow-V's exact prompt wrapper (fps -> noise-level -> aspect-ratio, in that order), and also encodes the empty-string unconditional caption through the *same* wrapper (with the same fps/noise_std/aspect_ratio metadata), matching upstream sample.py's `y += [""] * len(y)` performed before encode_text.

STARFlow V · 2 inputs · 3 parameters · 1 output

STARFlowVTextEncoderLoader

Loads a STARFlow-V text encoder from a selected local directory using the selected dtype.

STARFlow V · 1 output