Nodes/H3 Continuum/H3 Continuum Sampler V3.9
ComfyUI Node

H3 Continuum Sampler V3.9

H3 Continuum V3.9. Connect Reference Images V3.9 to assign fixed images 1–9 to all chunks or selected chunks.

By ukr8b3g-cmyk·Created 2 months ago·Updated about 15 hours ago· 113
H3 Continuum Sampler V3.9
  • model
  • clip
  • video_vae
  • sampler
  • sigmas
  • first_frame
  • last_frame
  • reference_video_1
  • driving_audio
  • audio_vae
  • reference_audio_1
  • reference_audio_vae
  • guide
  • audio_references
  • reference_images
  • video_latents
  • audio_latents
  • assembly_plan
  • status
  • driving_audio
  • refine_context
◄sequence_prompt—►
◄prompt_modeAuto►
◄chunks3►
◄chunk_seconds5.0►
◄aspectAuto from First Image►
◄presetDraft — 0.30 MP►
◄custom_mp0.30►
◄continuityBalanced — 22 frames►
◄base_seed0►
◄audio_continuitytrue►
◄diagnosticsBasic►
◄reroll_from_chunkAuto►
◄reroll_nonce0►
◄strict_compatibilitytrue►
◄debugfalse►
◄show_previewtrue►
◄run_storageOff►
◄run_name►
◄reference_sizeMatch Output►
◄project_id►
◄video_reference_sizeEfficient - 0.4 MP►
◄continuation_backendStandard►
◄generation_modeFull Run►
◄review_actionContinue / Next►
◄take_group0►
◄take_revision_id►
◄take_actionAutomatic►
◄size_sourceLegacy Aspect►
◄width736►
◄height416►
◄video_reference_modeRepeat Reference►
CategoryMiniMax H3/Continuum

Inputs (46)

NameTypeDefaultDescription
modelMODELMiniMax H3 diffusion model used for every Continuum chunk.
clipCLIPMiniMax H3 text encoder used to encode the complete Sequence Prompt.
video_vaeVAEUsed only to encode image conditioning. T2VA does not use it; Continuum never decodes with it.
samplerSAMPLERComfyUI sampler algorithm used unchanged for every generated chunk.
sigmasSIGMASComfyUI noise schedule used unchanged for every generated chunk.
sequence_promptSTRINGConnect one Text (Multiline) for the complete sequence.
prompt_modeCOMBOAutoAuto accepts Fixed, list-separated, and timeline prompt styles.
chunksINT31–16Number of sequential Continuum chunks to generate.
chunk_secondsFLOAT5.04–30Target duration shared by every chunk. 5–15 seconds is the recommended and validated range. Values above 15 seconds are supported, but VRAM use and processing time can increase substantially, especially at high resolution.
aspectCOMBOAuto from First ImageSaved V3.8 workflow compatibility only. The frontend migrates this value to Size Source and hides it from the Main UI.
presetCOMBODraft — 0.30 MPUsed only with Size Source = First Image. Draft is fastest; Balanced retains more detail; Native 768 uses the H3 native short edge; Custom uses Custom MP.
custom_mpFLOAT0.300.1–16Custom target megapixels while preserving the connected First Image aspect ratio. Used only when Resolution Preset is Custom.
continuityCOMBOBalanced — 22 framesAmount of prior video context retained at each chunk boundary.
base_seedINT00–18446744073709550000Base seed used to derive deterministic per-chunk seeds.
audio_continuityBOOLEANtrueOn passes prior audio context into continuation chunks. Turn it off only to isolate or replace generated audio.
diagnosticsCOMBOBasicControls the detail level of the read-only Status report. It does not change generated tensors.
reroll_from_chunkCOMBOAutoAuto resumes the longest compatible saved prefix. Choosing a chunk reuses earlier chunks and regenerates that chunk and everything after it.
reroll_nonceINT00–4294967295Change only when regenerating an explicit chunk and you want a new variation with otherwise identical settings.
strict_compatibilityBOOLEANtrueLegacy saved-workflow input. V3.8 keeps it loadable but ignores its value.
debugBOOLEANfalseDeveloper diagnostics controlled by the H3 Continuum settings panel.
show_previewBOOLEANtrueShow live sampling previews. Disable only to reduce preview overhead.
run_storageCOMBOOffAtomically save raw AV chunks and resume a compatible saved run.
run_nameSTRINGEnter a stable name for this saved run. Compatible chunks are selected automatically.
reference_sizeCOMBOMatch OutputMatch Output is the practical default; Max Identity preserves more reference detail.
project_idSTRINGOptional. Leave blank to derive a stable ID from this sampler node. Run Name remains the explicit override.
video_reference_sizeCOMBOEfficient - 0.4 MPEfficient limits Video Guide Frames to about 0.4 MP; Balanced uses about 0.6 MP; Match Output uses the output pixel area. Source aspect ratio is preserved and smaller sources are not enlarged.
continuation_backendCOMBOStandardStandard uses the V3.6 target-preserving continuation path. With Audio Continuity enabled, Masked AV currently uses Balanced 22; Fast 5, Strong 39, and Auto safely use Reference Context. Compatibility restores the V3.5 Reference Context path for older workflows or comparison. Run Storage identity follows the resolved transport before execution begins.
generation_modeCOMBOFull RunFull Run preserves normal Production execution. Review Each Chunk generates at most one new physical group per Queue and requires Run Storage = Save + Auto Resume.
review_actionCOMBOContinue / NextContinue / Next accepts the current review and advances. Regenerate Current and Finish Remaining are one-shot actions. Smart Regenerate requires Regenerate From = Auto.
take_groupINT00–16Internal Render History selection. Use the visible Previous/Next Take controls instead of editing this value directly.
take_revision_idSTRINGImmutable revision selected by Render History. The visible Take controls manage this value from verified Run Storage data.
take_actionCOMBOAutomaticOne-shot Render History action. Selecting a Take alone does not change the canonical branch; Queue normally after choosing an action.
size_sourceCOMBOLegacy AspectFirst Image preserves its aspect at the selected Resolution Preset. Manual uses Width and Height exactly. Legacy Aspect is accepted only for saved-workflow and API compatibility.
widthINT73632–16384Exact output width in Manual mode. Use a multiple of 32. Manual mode is the normal choice for T2VA or workflows without a First Image.
heightINT41632–16384Exact output height in Manual mode. Use a multiple of 32. Manual mode is the normal choice for T2VA or workflows without a First Image.
first_frameoptIMAGEFirst Image conditioning for I2VA or FL2VA. With Output Size = First Image, its aspect ratio also defines the output canvas. Leave it disconnected and use Manual Width/Height for T2VA.
last_frameoptIMAGEOptional last-frame anchor for FL2VA. Leave disconnected for T2VA and normal I2VA.
reference_video_1optIMAGE24 fps IMAGE frames from the Video Adapter. Follow Timeline selects successive physical output intervals; Repeat Reference keeps the legacy shared-prefix guide. Source audio is not included. Follow is Experimental; this is not a hard motion-copy control.
driving_audiooptAUDIOOptional original audio timeline. It is used as native H3 guide conditioning and selected unchanged for final output.
audio_vaeoptVAERequired only when Driving Audio is connected. Uses the same Audio VAE encode path as ComfyUI Core MiniMax H3 Add Guide.
reference_audio_1optAUDIOOptional standalone audio reference for H3 conditioning. It is not the audio track of Video Guide Frames. Unlike Driving Audio, it does not replace the generated final audio.
reference_audio_vaeoptVAERequired only when Reference Audio is connected. It encodes the reference for H3 conditioning; generated audio remains the output.
guideoptH3_CONTINUUM_STILL_IMAGE_GUIDEOptional V3.7 Still Image Guide. Only the owning physical sampling group receives the Core minimax_keyframes entry.
audio_referencesoptH3_CONTINUUM_AUDIO_REFERENCESOptional ordered bundle from H3 Continuum Reference Audios. Do not connect it together with the legacy single Reference Audio input.
video_reference_modeoptCOMBORepeat ReferenceExperimental. Follow Timeline selects the source interval for each physical generation group, then continues without video reference after the source ends. Repeat Reference preserves the previous Video Guide behavior: the same bounded prefix for every group. This is independent of Prompt Format. Input frames must be 24 fps.
reference_imagesoptH3_CONTINUUM_REFERENCE_IMAGES_V39Connect H3 Continuum Reference Images V3.9; assign chunks there.

Outputs (6)

NameTypeDescription
video_latentsLATENT—
audio_latentsLATENT—
assembly_planH3_CONTINUUM_ASSEMBLY_PLAN—
statusSTRING—
driving_audioAUDIO—
refine_contextH3_CONTINUUM_REFINE_CONTEXT—