This trainer output begins with Pruna/Replicate's REPLORA1 container signature. Despite the .safetensors filename, it is not a generic Safetensors file and will fail in generic Safetensors loaders. Use it with prunaai/p-image-lora as intended.

Training

text

Checkpoint:      sole step-1000 checkpoint from the owner-run pilot
Trigger:         ezflat_v1
Training type:   style
Steps:           1000
Learning rate:   0.0001
Default caption: unset
Images:          100
Captions:        100, one matching caption per image

Dataset composition:

text

plain-background characters
characters in scenes
close-up portraits
dynamic action poses
pairs/groups

Image dimensions were retained at 888×1184, 1024×1024, and 1184×888. Embedded ComfyUI metadata was stripped before training.

Source and status

The training targets were generated with Krea 2 Turbo and Krea2FlatIllustrationV1.safetensors at scale 1.0. The exact teacher file matches Civitai model version 3086495 by SHA-256. Civitai currently marks that source as allowing generated-image commercial use, derivatives, and different licensing.

Source page: https://civitai.com/models/2744176?modelVersionId=3086495

The teacher's author states that its own training images were generated using an artist tag. This repository is therefore kept private while the result is evaluated. No public release or production suitability is asserted here.

This checkpoint has not yet been visually benchmarked. Record the winning inference scale and prompt recipe after evaluation rather than treating 1.0 as final.

Checksums

text

weights.safetensors
ff7488ca3f227c361ec02ac58ad275958c217ee56cbe88f3ef98082f409bf019

oshi-pog/p-image-ezflat-v1

EZ Flat v1 — P-Image LoRA

Trigger

Compatibility

Training

Source and status

Checksums

oshi-pog/
p-image-ezflat-v1