Nodes/ComfyUI_MieNodes/MiniMax H3 Prompt Generator ๐Ÿ‘
ComfyUI Node

MiniMax H3 Prompt Generator ๐Ÿ‘

A ComfyUI node in ๐Ÿ‘ MieNodes/๐Ÿ‘ Prompt Generator with 19 inputs and 1 output.

By MieMieeeeeยทCreated 2 years agoยทUpdated 6 days agoยท 235
MiniMax H3 Prompt Generator ๐Ÿ‘
  • llm_service_connector
  • first_frame
  • last_frame
  • reference_images
  • reference_video
  • h3_prompt
โ—„task_typet2v - ๆ–‡็”Ÿ่ง†้ข‘โ–บ
โ—„user_promptโ–บ
โ—„seed0โ–บ
โ—„width1280โ–บ
โ—„height720โ–บ
โ—„duration6.0โ–บ
โ—„categorynone - ไธๆŒ‡ๅฎšโ–บ
โ—„output_languageenโ–บ
โ—„caption_sample_frames8โ–บ
โ—„image_detailautoโ–บ
โ—„temperature0.40โ–บ
โ—„max_tokens_caption4096โ–บ
โ—„max_tokens_enhance8192โ–บ
โ—„timeout120โ–บ
Category๐Ÿ‘ MieNodes/๐Ÿ‘ Prompt Generator

Inputs (19)

NameTypeDefaultDescription
llm_service_connectorLLMServiceConnectorโ€”
task_typeCOMBOt2v - ๆ–‡็”Ÿ่ง†้ข‘6 options: t2v - ๆ–‡็”Ÿ่ง†้ข‘, i2v_first - ๅ›พ็”Ÿ่ง†้ข‘(้ฆ–ๅธง), i2v_first_last - ๅ›พ็”Ÿ่ง†้ข‘(้ฆ–ๅฐพๅธง), i2v_last - ๅ›พ็”Ÿ่ง†้ข‘(ๅฐพๅธง), reference - ๅ…จ่ƒฝๅ‚่€ƒ, s2v - ไธปไฝ“ๅ‚่€ƒ
user_promptSTRINGโ€”
seedINT00โ€“18446744073709550000โ€”
first_frameoptIMAGEโ€”
last_frameoptIMAGEโ€”
reference_imagesoptIMAGEโ€”
reference_videooptIMAGEReference video for the 'reference' task, supplied as an IMAGE batch (one tensor per frame). caption_sample_frames of them are sampled evenly for the stage-1 caption. Audio references are not wired (ComfyUI IMAGE port carries no audio); describe audio inside user_prompt instead.
widthoptINT1280Output frame width in pixels. Together with height it defines the aspect ratio written into the prompt header (e.g. 1280x720 -> '16:9'). The downstream H3 video node sets the real render size.
heightoptINT720Output frame height in pixels. Together with width it defines the aspect ratio written into the prompt header (e.g. 720x1280 -> '9:16').
durationoptFLOAT6.0Target video duration in seconds (float). Written into the prompt header and the FL2VA/L2VA final-timestamp directive. NOTE: the downstream H3 video node only accepts 5/6/8/12/15s; other values will make the video node error, not this one.
categoryoptCOMBOnone - ไธๆŒ‡ๅฎšH3 use-case category. Adds default styling advice to the prompt (e.g. cinematic-story -> cinematic color grading + 35-50mm; action -> motion blur + shake). Does not change the task path. 'none' = no category-specific advice.
output_languageoptCOMBOenOutput language of the final H3 prompt. 'en' = English (default), 'zh' = Chinese. The section headers (Core idea / Soundscape / Music / Do not include) switch to the chosen language; the descriptive body is written in that language too. EXCEPTION: dialogue, lyrics, and visible on-screen text always stay in their original language.
caption_sample_framesoptINT81โ€“16Number of frames sampled evenly from reference_video for the stage-1 caption (reference task only). Higher = more accurate caption but slower and costlier; lower = faster but may miss motion. Does NOT set the output video length.
image_detailoptCOMBOauto3 options: auto, low, high
temperatureoptFLOAT0.400โ€“2โ€”
max_tokens_captionoptINT409664โ€“32768โ€”
max_tokens_enhanceoptINT819264โ€“32768โ€”
timeoutoptCOMBO1204 options: 30, 60, 120, 300

Outputs (1)

NameTypeDescription
h3_promptSTRINGโ€”