FASHN v1.6: Our Best Virtual Try-On Model Yet, Now at 864 x 1296 Resolution

What's New in FASHN v1.6?

• Native 864x1296 Resolution (1MP) - FASHN v1.6 delivers results at a native resolution of 864x1296 pixels, representing a 50% increase compared to v1.5. The improved resolution ensures clearer details such as fabric textures and garment patterns without compromising the established accuracy of earlier versions.

• Improved Identity Preservation (Quality Mode) - The 'Quality' mode has been upgraded to better maintain original image integrity. Improvements include more accurate skin texture preservation, and smarter top & bottom interactions, resulting in more authentic virtual try-ons.

• Optimized Performance with Minimal Runtime Impact - Higher resolution typically means increased runtime; however, we've optimized the processing to offer enhanced results efficiently:

  • Performance Mode (~7 seconds): Provides improved visual detail compared to v1.5's 'Balanced' mode, with only a minor runtime increase (+1 second).

  • Balanced Mode (~10 seconds): Further refines garment detail and texture. More reliable in more challenging try-on tasks. 

  • Quality Mode (~19 seconds): Prioritizes overall image quality, extensively improving skin texture fidelity and realistic garment integration with the original model image.

Visual Comparisons: FASHN v1.5 vs. v1.6

Below are visual examples illustrating the improvements introduced in FASHN v1.6. We compare it directly with the previous version, FASHN v1.5, and demonstrate how FASHN v1.6 performs across its various runtime modes: Performance, Balanced, and Quality.

Example 1: Columbia Puffer Jacket (On-Model)

_Inputs:

_Comparison:

Initially, we processed the input images using FASHN v1.5 in Balanced mode, achieving a satisfactory result at a resolution of 576×864 pixels within approximately 6 seconds.

We then applied the identical inputs and seed using FASHN v1.6 in two modes: Performance (7 seconds) and Balanced (10 seconds).

FASHN v1.6 generated an output at 768×1152 pixels. This higher-resolution result is noticeably sharper, more accurate, and richer in detail compared to the previous v1.5 version.

_{Note: While FASHN v1.6 supports resolutions up to 864×1296 pixels, the example's output resolution (768×1152) matches the resolution of the input model images. FASHN v1.6 does not upscale beyond the resolution of the original input.}

Balanced mode in v1.6, taking an additional 3 seconds compared to Performance mode, delivers further refinement. This extra processing time notably enhances fine details and resolves minor imperfections, such as improved alignment between the zipper slider and zipper teeth.

Example 2: Midi Dress Flat-Lay to Plus Model

_Inputs:

_Comparison:

Consistent with the previous example, we first ran FASHN v1.5 in Balanced mode, resulting in a 576×864 output within 6 seconds.

When processed with FASHN v1.6, the output resolution increased to 768×1152 pixels, enhancing both detail accuracy and sharpness. Noticeably, v1.6 also produced more anatomically correct legs for the model.

The Balanced mode in v1.6 further sharpened and refined details, now perfectly replicating the graphic print from the original garment.

Additionally, we ran FASHN v1.6 in Quality mode (19 seconds) for this example, significantly improving realism by restoring the model’s natural skin tones on the arms and neck, and refining the anatomical accuracy of the legs for an even more realistic final result.

Example 3: H&M Suit Set (On-Model) to Fitting Room Selfie

_Inputs:

_Comparison:

Moving towards more realistic, user-generated content (UGC) style images, this example explores FASHN v1.6 performance on fitting room selfies in a virtual try-on scenario.

FASHN v1.6 Performance mode notably improves color accuracy, texture representation, and resolves minor details such as the white shirt collar and a previously missing finger. Balanced mode further refines lighting adjustments to better match the scene and removes small artifacts near the index finger, providing a more polished result.

Example 4: FASHION NOVA Gown (On-Model) to Plus Model

_Inputs:

_Comparison:

FASHN v1.6 in Performance mode displays significantly enhanced detail, more realistic lighting effects, and improved fabric transparency on the lower skirt.

Using the Balanced mode, FASHN v1.6 further corrects fine details, achieving a more faithful representation of the original gown.

Additionally, the Quality mode of FASHN v1.6 subtly but crucially restores the authentic appearance of the arms and neck areas, ensuring the highest level of realism and accuracy.

Example 5: Zara Polo Shirt (On-Model), Side View Try-On

_Inputs:

_Comparison:

FASHN v1.5 already did a great job fitting the polo shirt, but FASHN v1.6 Performance mode still shines with a higher resolution and more accurate, deeper texture transfer. 

Balanced mode highlights even more fine details. 

Quality mode pushes the try-on result to perfection by completely preserving the appearance of the model’s forearm and denim shorts.

Example 6: Baggy DIESEL Jeans (On-Model) to Tight Shorts

_Inputs:

_Comparison:

Like with the previous example, FASHN v1.5 already did a great job fitting these baggy jeans. However, FASHN v1.6 takes it a step further by capturing additional fine details, such as precise stitching, pocket definition, fabric folds, and enhanced denim texture, significantly elevating the overall realism and accuracy.

How to Access FASHN v1.6

Web App Users: FASHN v1.6 is now the default model powering our Try-On Studio.

API Users: Please read our updated API documentation to learn how to access FASHN v1.6.

What's Next?

We’re very proud of this release, but acutely aware that there are still many challenges we need to overcome to provide a perfect try-on solution. Our goals include even higher resolutions, faster runtimes, more complex fit handling, and perfect preservation of individual body shapes and unique characteristics. We remain dedicated to continuously improving the world's leading virtual try-on solution.

— The FASHN Team