Tensor.Art
Create

RefractiveXL - v1

RefractiveXL

CHECKPOINT
Original

Updated:

v1

RefractiveXL excels at photorealism. While some anime and other models were added to the merge, it is not the focus of this model.

Tested with:

Sampler - DPM++ 3M SDE

Schedule - Karras

Steps - 40

CFG - 7.5

Clip Skip - 1/no clip skip

Resolutions:

1:1 aspect ratio

•          768x768 (minimum recommendation)

•          1024x1024 (standard quality & performance)

•          1280x1280 (high quality, normal performance, most recommended)

•          1344x1344 (ultra-high quality, slow performance)

•          1532x1532 (not recommended, prone to stretching and character doubling)

9:16 aspect ratio

•          648x1152 (minimum recommendation)

•          720x1280 (high quality, normal performance, most recommended)

•          792x1408 (high quality, normal performance)

•          864x1536 (ultra-high quality, normal performance)

•          936x1664 (ultra-high quality, slow performance, not recommended*)

4:7 aspect ratio

•          672x1176 (minimum recommendation)

•          704x1232 (best blend of quality and speed, most recommended)

•          736x1288 (high quality, fast performance)

•          768x1344 (high quality, fast performance)

•          800 x 1400 (high quality, normal performance)

•          832 x 1456 (high quality, normal performance)

•          864 x 1512 (ultra-high quality, normal performance)

•          896x1568 (ultra-high quality, slow performance, not recommended)

 

*note: any time resolution is greater than or equal to 1536 in length or height, while it can produce amazing quality images, there is a potential chance to have a doubled or stretched image subject, as it is 2x the 768x768 training resolution generally used in some of the older SDXL models within this merge. 1:1 aspect ratio can be expected fair worse more often than 4:7 and 9:16.

Version Detail

SDXL 1.0
RefractiveXL is a merged checkpoint model which excels at photorealism. While some anime and other models were added to the merge, it is not the focus of this model and may not consistently produce anime styled images. Tested with: Sampler - DPM++ 3M SDE Schedule - Karras Steps - 40 CFG - 7.5 Clip Skip - 1/no clip skip, no noticeable difference with 2k+ characters when testing clip skip 2 thru 12. Resolutions: 1:1 aspect ratio • 768x768 (minimum recommendation) • 1024x1024 (standard quality & performance) • 1280x1280 (high quality, normal performance, most recommended) • 1344x1344 (ultra-high quality, slow performance) • 1532x1532 (not recommended, see notes below*) 9:16 aspect ratio • 648x1152 (minimum recommendation) • 720x1280 (high quality, normal performance, most recommended) • 792x1408 (high quality, normal performance) • 864x1536 (ultra-high quality, normal performance) • 936x1664 (ultra-high quality, slow performance, not recommended*) 4:7 aspect ratio • 672x1176 (minimum recommendation) • 704x1232 (best blend of quality and speed, most recommended when speed is the need) • 736x1288 (high quality, fast performance) • 768x1344 (high quality, fast performance) • 800 x 1400 (high quality, normal performance) • 832 x 1456 (high quality, normal performance) • 864 x 1512 (ultra-high quality, normal performance, highest quality recommended) *note: any time resolution is greater than or equal to 1536 in length or height, while it can produce amazing quality images, there is a potential chance to have a doubled or stretched image subject, as it is 2x the 768x768 training resolution generally used in some of the older SDXL models within this merge. 1:1 aspect ratio can be expected fair worse more often than 4:7 and 9:16.

Project Permissions

    Use Permissions

  • Use in TENSOR Online

  • As a online training base model on TENSOR

  • Use without crediting me

  • Share merges of this model

  • Use different permissions on merges

    Commercial Use

  • Sell generated contents

  • Use on generation services

  • Sell this model or merges

Related Posts