Nodes/H3 Continuum/H3 Continuum Hi-Res Fix V3.5
ComfyUI Node

H3 Continuum Hi-Res Fix V3.5

Continuum-aware Hi-res fix. When enabled, each first-pass physical video LATENT is decoded with the connected H3 Video VAE, resized in pixel space, re-encoded with the same VAE, and refined once with the workflow SIGMAS. scale_by 0 uses ComfyUI Core's H3 native canvas; positive values use the manual multiplier. Connect the optional Continuum refine context to preserve First/Last/Reference conditioning at the target geometry. When disabled, first-pass video/audio LATENT objects and the original assembly plan are returned unchanged.

By ukr8b3g-cmyk·Created 15 days ago·Updated about 9 hours ago· 21
H3 Continuum Hi-Res Fix V3.5
  • model
  • clip
  • sampler
  • sigmas
  • video_latents
  • audio_latents
  • assembly_plan
  • refine_context
  • video_vae
  • video_latents
  • audio_latents
  • updated_assembly_plan
  • status
enabledtrue
upscale_methodlanczos
scale_by2.00
refine_seed0
CategoryMiniMax H3/Continuum

Inputs (13)

NameTypeDefaultDescription
modelMODEL
clipCLIP
samplerSAMPLER
sigmasSIGMAS
video_latentsLATENT
audio_latentsLATENT
assembly_planH3_CONTINUUM_ASSEMBLY_PLAN
enabledBOOLEANtrueEnable Resize + Second Pass. Off returns the first-pass video/audio LATENT objects and assembly plan unchanged.
upscale_methodCOMBOlanczos6 options: lanczos, nearest-exact, bilinear, area, bicubic, bislerp
scale_byFLOAT2.000–42.0 = validated 2x Pixel/VAE Hi-res Fix. 0 uses the ComfyUI Core H3 native canvas automatically. Manual values follow the existing Second Pass contract and must be 1.0 or greater.
refine_seedINT00–18446744073709550000Base seed independently derived once per physical group.
refine_contextoptH3_CONTINUUM_REFINE_CONTEXT
video_vaeoptVAERequired while enabled for the safe Pixel/VAE roundtrip; not evaluated while disabled.

Outputs (4)

NameTypeDescription
video_latentsLATENT
audio_latentsLATENT
updated_assembly_planH3_CONTINUUM_ASSEMBLY_PLAN
statusSTRING