Nodes/Comfy_HunyuanImage3/Hunyuan Instruct Multi-Image Fusion
ComfyUI Node

Hunyuan Instruct Multi-Image Fusion

A ComfyUI node in Hunyuan/Instruct with 16 inputs and 3 outputs.

By EricRollei·Created 9 months ago·Updated 3 months ago· 64
Hunyuan Instruct Multi-Image Fusion
  • model
  • image_1
  • image_2
  • image_3
  • image_4
  • image_5
  • image
  • cot_reasoning
  • status
instructionCombine the style of image 1 with elements from image 2
seed-1
bot_taskimage
system_promptdynamic
resolution1024x1024 (1:1 Square)
steps-1
guidance_scale-1.0
flow_shift2.80
max_new_tokens2048
verbose0
CategoryHunyuan/Instruct

Inputs (16)

NameTypeDefaultDescription
modelHUNYUAN_INSTRUCT_MODEL
image_1IMAGE
instructionSTRINGCombine the style of image 1 with elements from image 2Instruction describing how to combine the images. Reference images as 'image 1', 'image 2', 'image 3', etc. Examples: 'Based on image 1 logo, create a fridge magnet with image 2 material', 'Let the cat from image 1 take a selfie with the cat from image 2, with image 3 as background'. Works with both English and Chinese prompts.
seedINT-1-1–2147483647-1 for random seed
image_2optIMAGE
image_3optIMAGE
image_4optIMAGEExperimental — model officially supports up to 3, but the pipeline accepts more. Increases VRAM usage significantly.
image_5optIMAGEExperimental — model officially supports up to 3, but the pipeline accepts more. Increases VRAM usage significantly.
bot_taskoptCOMBOimageControls how the model processes your fusion instruction.\n• image: Direct fusion — instruction applied as-is.\n• recaption: Rewrites instruction into detailed fusion description then generates.\n• think_recaption: (BEST) CoT reasoning about how to combine elements, rewrites instruction, then generates. Recommended for complex multi-image fusion.
system_promptoptCOMBOdynamicSystem prompt guiding multi-image fusion. dynamic (recommended) auto-selects the best prompt for your bot_task. en_unified includes reference editing guidelines for extracting and combining elements across images. The model understands both English and Chinese instructions.
resolutionoptCOMBO1024x1024 (1:1 Square)Output image resolution. Auto lets the model predict aspect ratio (often unpredictable with multiple input images of different sizes). Selecting a specific resolution gives deterministic output dimensions.
stepsoptINT-1-1–100-1 for auto (8 for Distil, 50 for full Instruct). Higher step counts (60–80) reduce flow-matching artifacts at 2K+ resolutions but generation time scales linearly.
guidance_scaleoptFLOAT-1.0-1–20CFG scale. -1 = auto (model's recommended value, typically 2.5)
flow_shiftoptFLOAT2.800–10Flow shift for the diffusion scheduler. Controls denoising schedule shape. Default 2.8 is balanced. Presets: Portraits / faces: 2.0–2.5 (sharper detail). Landscapes / illustrations: 3.5–5.0 (cleaner gradients).
max_new_tokensoptINT2048256–8192Maximum tokens for CoT reasoning and instruction rewriting. Only used with recaption or think_recaption modes. Multi-image fusion may benefit from higher values (e.g. 3072) for complex instructions.
verboseoptINT00–2Verbosity level. 0=silent (recommended), 1=info (shows full system prompt), 2=debug

Outputs (3)

NameTypeDescription
imageIMAGE
cot_reasoningSTRING
statusSTRING