Nodes/ComfyUI-Minimax-H3-Prompt-Engineer/MiniMax H3 Prompt Studio + Generate
ComfyUI Node

MiniMax H3 Prompt Studio + Generate

One-node prompt engineering and official MiniMax H3 conditioning. Connect each image, video, or audio once; use @图像1, @视频1, @视频音频1, and @音频1 in the request.

By colorAi·Created 22 days ago·Updated 19 days ago· 19
MiniMax H3 Prompt Studio + Generate
  • clip
  • vae
  • audio_vae
  • first_frame
  • last_frame
  • ref_images
  • ref_videos
  • ref_video_audios
  • ref_audios
  • creative_presets
  • positive
  • latent
  • formatted_prompt
  • validation_report
  • raw_response
  • usage_json
  • display_prompt
ai_provider
prompt_templateOfficial H3 · General
request_levelBasic · 基础
display_languageEnglish · H3 native
task_modeAuto · Infer from connected assets
user_request
width1344
height768
length124
ref_image_sizematch
reference_context
temperature0.20
top_p0.90
max_tokens4096
reasoning_effortnone
timeout_seconds120
image_max_side1024
auto_repairtrue
strict_validationtrue
CategoryMiniMax H3/Prompt Engineer

Inputs (29)

NameTypeDefaultDescription
clipCLIP
vaeVAE
ai_providerCOMBOUse prompt as-is skips all LLM calls and H3 document validation. Prompt already formatted keeps strict schema validation.
prompt_templateCOMBOOfficial H3 · General9 options: Official H3 · General, 3D Animation Short · 3D 动画短片, Brand Promo · 品牌宣传片, Co-op Game Intro · 合作游戏开场, Hand-drawn Live · 手绘实拍融合, Minimalist Product Ad · 极简产品广告, +3
request_levelCOMBOBasic · 基础Basic uses a compact H3 contract and short output; Medium adds production rules; Full injects the complete official writing guides.
display_languageCOMBOEnglish · H3 nativeAI providers always send English to H3 and can add a Chinese display translation. Direct modes return the supplied language unchanged.
task_modeCOMBOAuto · Infer from connected assets6 options: Auto · Infer from connected assets, T2VA · Text to Audiovisual, I2VA · First Frame to Audiovisual, FL2VA · First and Last Frames to Audiovisual, L2VA · Last Frame to Audiovisual, FULL_REFERENCE · Full Reference
user_requestSTRING
widthINT134432–16384
heightINT76832–16384
lengthINT1245–3600Frame count at 24 fps; H3 snaps upward to its 17k+5 frame grid.
ref_image_sizeCOMBOmatchMatches the official MiniMax H3 Reference to Video node.
reference_contextSTRING
temperatureFLOAT0.200–2
top_pFLOAT0.900–1
max_tokensINT4096512–32768
reasoning_effortCOMBOnone7 options: none, minimal, low, medium, high, xhigh, +1
timeout_secondsINT12010–600
image_max_sideINT1024512–4096
auto_repairBOOLEANtrue
strict_validationBOOLEANtrueApplies to AI output and strict Direct documents; pass-through Direct always skips it.
audio_vaeoptVAE
first_frameoptIMAGE
last_frameoptIMAGE
ref_imagesoptCOMFY_AUTOGROW_V3
ref_videosoptCOMFY_AUTOGROW_V3
ref_video_audiosoptCOMFY_AUTOGROW_V3
ref_audiosoptCOMFY_AUTOGROW_V3
creative_presetsoptH3_CREATIVE_PRESET

Outputs (7)

NameTypeDescription
positiveCONDITIONING
latentLATENT
formatted_promptSTRING
validation_reportSTRING
raw_responseSTRING
usage_jsonSTRING
display_promptSTRING