Description
The model makes 128x128 (upscaled to 1024x1024) or 160x160 (1280x1280) images on Krea2 Turbo with 8x8 logical pixel size (INT8 ConvRot, INT4, GGUF formats stay usable)
Use the 'Pixel Grid: Median Fixer' node with 'Grid size 8' to post-process the image for absolute
https://registry.comfy.org/nodes/comfyui-pixelgrid-helpers
Combining the LoRA with other LoRAs keep the 8x8 pixel sizes, but its best if other LoRAs are given a slightly lower strength.
Workflow:
Prompting guide
Objective
Write a direct, factual caption describing what is clearly visible in the image. Avoid guessing identities, emotions, background stories, or artistic mediums.
Description Order
Foreground Subjects: List main people, animals, or prominent objects in order of visual impact. Separate each subject with a semicolon.
Background: Add the label
Background:immediately after the subjects, then describe the environment, key surroundings, arrangement, and lighting.
Formatting Template
<subject 1 description>; <subject 2 description>. Background: <setting, surroundings, arrangement, and lighting>
(If the image lacks a clear main subject, start directly with No distinct subject. followed by Background: ...)
Guidelines
Focus on distinct physical traits, clothing, poses, or actions to identify subjects.
Describe the scene clearly without listing every minor detail or repeating location terms.
Keep the output to a single plain-text paragraph with no extra labels, quotes, or markdown
