metadata

language:
  - en
tags:
  - stable-diffusion
  - text-to-image
license: creativeml-openrail-m
inference: false

ReFT: Fine Tuned models by AI Generated Images

ReFT is a series of fine-tuned models from SD 1.5 for producing illustration style images with high resolution or different aspect ratio images.

Model Description

Model Name	Pretrained Model	Typical Resolusion
ReFT Rellust 512p	SD1.5	512x512
ReFT Rellust 768p	Rellust_512p_e10	768x768
ReFT Rellust 1K	Rellust_768p_e3	1024x1024

CLIP Skip 1 is recommended for all models!!

ReFT Rellust

Each model is fine-tuned from SD1.5 with ~16k nijijourneyv5 tagged images of various authors published on the Internet with a learning rate of 4e-6 using Aspect Ratio Bucketing based on either 512p, 768p, or 1K resolution. Image captions are made by BLIP and WD1.4-tagger. The batch size is 16 for 512p only and 8 for the rest. Fine tuning is also performed at fp16, using mutires noise.

Usage

Since these models are generic illustration models, the generated images can be in a variety of styles. If the style is not to your liking, please specify a style such as "anime" or "wator colors" as appropriate.

The format of the caption suggests that a short natural language sentence followed by a comma-separated tags is the most natural way to describe the prompt. Using more tags that are well-estimated by the tagger in the trained images may lead to more preferable generation. Tag semantics may be inappropriate for automatic tagging, so please emphasize appropriately. The models do not require underscores for tags in the prompts.

CLIP Skip 1 is recommended.

Examples (ReFT Rellust 1K)

solo, 1girl, full body, white background, indoors, from side, short hair, bangs, long sleeves, closed mouth, white hair, jewelry, blush, standing, earrings dress, flower, shirt, blue eyes, hair ornament, skirt, bow
Negative prompt: worst quality, low quality, bad anatomy blurry, text
Steps: 30, Sampler: UniPC, CFG scale: 7, Seed: 4203630207, Size: 1280x720, Model hash: 7d610c9f1e, Model: Rellust_1K_e3, Denoising strength: 0.55, FreeU Stages: "[{\"backbone_factor\": 1.2, \"skip_factor\": 0.9}, {\"backbone_factor\": 1.4, \"skip_factor\": 0.2}]", CFG Rescale phi: 0, Hires upscale: 1.5, Hires steps: 20, Hires upscaler: Latent, Version: v1.5.1

wide shot, cosmos, white background, simple background, full body, solo, 1girl, small head, light-brown hair, medium hair, wavy hair, brown eyes, simple dress, casual, slim, strap shoes, cosmos, water colors, pastel colors, light coloring
Negative prompt: worst quality, low quality, logo, blurry, text, frame, light particle, water, vivid color, black, messy hair, frills
Steps: 30, Sampler: UniPC, CFG scale: 6, Seed: 122211893, Size: 1024x1024, Model hash: 7d610c9f1e, Model: Rellust_1K_e3, Denoising strength: 0.6, FreeU Stages: "[{\"backbone_factor\": 1.2, \"skip_factor\": 0.9}, {\"backbone_factor\": 1.4, \"skip_factor\": 0.2}]", CFG Rescale phi: 0, Hires upscale: 1.5, Hires steps: 20, Hires upscaler: Latent, Version: v1.5.1

wide shot, grapes, white background, simple background, full body, solo, 1girl, small head, light-brown hair, medium hair, wavy hair, brown eyes, simple dress, casual, slim, shoes, muscat, water colors, pastel colors, light coloring, grapes
Negative prompt: worst quality, low quality, lowres, blurry, text, frame, light particle, water, vivid color, black, messy hair, frills
Steps: 30, Sampler: UniPC, CFG scale: 7, Seed: 3764144276, Size: 1024x1024, Model hash: 7d610c9f1e, Model: Rellust_1K_e3, Denoising strength: 0.6, FreeU Stages: "[{\"backbone_factor\": 1.2, \"skip_factor\": 0.9}, {\"backbone_factor\": 1.4, \"skip_factor\": 0.2}]", CFG Rescale phi: 0, Hires upscale: 1.5, Hires steps: 20, Hires upscaler: Latent, Version: v1.5.1

License

This model is open access and available to all, with a CreativeML OpenRAIL-M license further specifying rights and usage. The CreativeML OpenRAIL License specifies:

You can't use the model to deliberately produce nor share illegal or harmful outputs or content
The authors claims no rights on the outputs you generate, you are free to use them and are accountable for their use which must not go against the provisions set in the license
You may re-distribute the weights and use the model commercially and/or as a service. If you do, please be aware you have to include the same use restrictions as the ones in the license and share a copy of the CreativeML OpenRAIL-M to all your users (please read the license entirely and carefully) Please read the full license here

Acknowledgements

These Models build on the excellent work: SD1.5, developed by CompVis and RUNWAY