Vision Realistic DiT

Vision Realistic Model Overview

I’m excited to introduce my latest checkpoint model, based on HunyuanDiT-v1.2. This model has been trained over 80,000 steps to ensure the generation of high-quality, photorealistic images.

Model Details :

• Type: Photorealistic model

• Trigger Words: None required

• Chinese language support: No

• Output: High-detail, high-resolution images that closely resemble real-life photographs

Configuration Used for Training:

• GPU: A6000

• Dataset: Combination of 5,000 stock photos and my own custom dataset

• Batch Size: 2

• Optimizer: AdamW

• Scheduler: Cosine

• Learning Rate: 1e-5

• Epochs: Target of 100 epochs

• Captioning: Mixed WD14 and BLIP

• Training Time: 27+ hours (Experience: Bad; future training undecided)

Quick Guide and Parameters:

• VAE: SDXL

• Sampler: dpmpp_2m

• Scheduler: sgm_uniform (Recommended for best results)

• Sampling Steps: 25+

• CFG Scale: 7

For better results, try using ComfyUI

• workflow

Important: Please avoid using NSFW/mature content in your prompts, as it may lead to unreliable results. Additionally, shorter prompts tend to work better with both SD3 and DiT models.

Note:

This is not a merged or modified model. It is the original Realistic Vision fine-tuned model. Some users have been spreading incorrect information in the model's comment section. If you have any questions or want to know more, join my Discord server or share your thoughts in the comment section. Thank you for your time.