Files
tenser-tensor-ComfyUI-Tense…/web/docs/TT_LatentFactory.md
T

2.7 KiB
Raw Blame History

TT Latent Factory TT_LatentFactory

This description was generated with AI assistance. If you spot any errors, please create an issue on GitHub.

Generates empty latent images with precise dimension control based on aspect ratio and megapixel count. Automatically calculates correct dimensions for FLUX and SDXL models, plus separate CLIP conditioning sizes.

Inputs

Parameter Type Required Description
seed INT Yes Seed for reproducible results (0 – 18446744073709551615)
noise_seed INT Yes Separate seed for noise generation (0 – 18446744073709551615)
aspect_ratio STRING Yes Image proportions (1:1, 4:3, 3:2, 16:9, 21:9)
megapixels COMBO Yes Total resolution (0.25 MP, 0.5 MP, 1 MP, 2 MP, 4 MP, 8 MP)
orientation STRING Yes Image orientation (landscape / portrait)
model_type STRING Yes Target model architecture (FLUX1.D, FLUX2.D, SDXL)
batch_size INT Yes Number of latents to generate (1–64, default: 1) - advanced
clip_multiplier STRING Yes CLIP target size multiplier (1x, 2x, 4x) - advanced

Outputs

Parameter Type Description
LATENT LATENT Empty latent tensor ready for sampling
RANDOM_NOISE NOISE Noise object seeded with noise_seed
SEED INT Generation seed
NOISE_SEED INT Noise seed
MEGAPIXELS COMBO Selected megapixel value for chaining
WIDTH INT Calculated image width (multiple of 64)
HEIGHT INT Calculated image height (multiple of 64)
TARGET_WIDTH INT CLIP conditioning target width
TARGET_HEIGHT INT CLIP conditioning target height

Usage

Select aspect ratio, resolution, and orientation to automatically compute pixel dimensions suited for the chosen model. FLUX1.D uses 16-channel latents, FLUX2.D uses 128-channel, SDXL uses 4-channel. clip_multiplier scales the TARGET dimensions relative to the image dimensions — useful for multi-resolution CLIP conditioning. Connect RANDOM_NOISE directly to a sampler node for reproducible noise tied to noise_seed.


View on GitHub