Introduction

As FASHN grows and welcomes more users, your feedback and insights play a crucial role in shaping our priorities.

Virtual try-on technology for clothing enables a wide range of applications, and we believe that sharing our plans transparently will help you understand if our direction is aligned with your vision.

Here’s a snapshot of where we’re headed, driven by your needs and ideas.

Virtual Try-On Model

At the heart of our platform is our unique virtual try-on AI model. Here’s what’s coming next:

1. Higher Output Resolution

The new 576x864 resolution is now available for testing via our web app and API, and the feedback has been overwhelmingly positive. However, one challenge we’re addressing is its runtime, which is nearly twice as long as the 384x576 model. We are actively optimizing this higher-resolution model, aiming to reduce its runtime to match the current 8–11 seconds. Once achieved, this will become our primary model.

2. Lightning-Fast Runtime

We’re also developing a hyper-optimized version of the model, targeting a nearly instant runtime of just 2–3 seconds. This version may involve a slight trade-off in accuracy, but it’s designed for scenarios where speed is critical.

3. Remove Bias from Starting Garment

You may have noticed for example that fitting a t-shirt on someone wearing a suit or coat doesn’t always yield the best results. This happens because the initial clothing creates a volume bias, influencing the AI’s interpretation of what constitutes a good fit. We’ve developed promising solutions to address this and will begin applying them to the 576x864 model once it’s fully stabilized.

4. Better Identity Preservation

Our virtual try-on pipeline occasionally struggles to retain the original characteristics of the model image, leading to unintended changes such as altering body proportions or adding/removing features like tattoos. We’re confident in our ability to improve this aspect and plan to implement these enhancements alongside the garment bias fixes, ensuring greater consistency and accuracy in the results.

5. Prompt Styling

We will introduce the ability to add subtle styling hints to your try-ons, such as “open/closed jacket,” “rolled-up sleeves,” or “tucked-in shirt.” This feature will allow you to have greater control over garments that can be styled in multiple ways, offering a more personalized and flexible virtual try-on experience.

6. Training Data Collection

Even the most advanced architectural improvements still rely on high-quality training data to perform at their best. Based on your feedback, we’re continuing to expand our dataset with a focus on these key areas:

Flat-Lay Photography: Ensuring the model can accurately interpret garments presented in flat-lay or ghost mannequin formats.
User-Generated Content (UGC): Incorporating amateur-quality photos to enhance the model’s robustness across diverse real-world scenarios.
Indian Ethnic Wear: Expanding support for traditional clothing styles to better serve users with unique cultural and regional preferences.

While we’re excited about the features we’re working on, we believe it’s equally important to share what’s not currently in our plans, even if they’re highly requested. Here’s what’s on hold for now:

Clothes Layering

Combining layers—such as fitting an open jacket over a specific t-shirt.

Fitting Top & Bottom Together

Supporting the simultaneous fitting of two clothing items (e.g., a top and a bottom) isn’t in our immediate plans. However, feature (4) of the virtual try-on model roadmap will greatly improve the ability to fit items consecutively while preserving the details of the first item.

Non-Clothing Categories

Try-ons for other categories like shoes, hats, jewelry, or accessories are not something we anticipate supporting anytime soon.

Web App

Alongside the core improvements to our try-on model, we aim to introduce new features and tools to enhance the FASHN app’s user interface:

1. Mockup Generator

To close the gap between clothing design and AI-powered photoshoots, we’re building a user-friendly interface for generating mockups directly within the FASHN app. This feature will let you instantly visualize your designs on pre-made garment and on-model templates, which you can then move to the try-on studio for more model and pose variations.

2. Model Management

Whether you’re using your own model photos or AI-generated models from our platform, we aim to make managing them easier. The FASHN app will soon allow you to save and organize your favorite model photos, ensuring quick access to AI generations you love or models that consistently deliver great try-on results.

3. Outpainting

For users who need resolutions beyond 576x864, we’re introducing a creative solution: outpainting. This feature will let you perform the try-on at 576x864 and then expand the image outward to a higher resolution using generative AI, maintaining a realistic and cohesive look.

There are additional features we’re planning to add that still require research and development before we’re ready to integrate them:

Background Change + Relighting

We’re exploring the ability to realistically change the background of an image, either before, after, or independent of the try-on process in the FASHN app. To ensure the final image looks natural, we’re also researching techniques to adjust the lighting so the background aligns seamlessly with the foreground (the person). Once we’re confident in the quality of this feature, we’ll work to bring it to the platform.

Model/Face Swap

An alternative to using virtual try-on to create more on-model image variations is to retain the clothing but replace the model’s face or body, essentially swapping everything around the garment. Current commercially available models don’t yet deliver the high-resolution, realistic results we aim for in this use case. However, we plan to integrate this feature as soon as advancements in text-to-image generation technology meet our quality standards.

Developer API

For our API users who are using FASHN to power their own apps, here are the features we expect to add soon to improve your experience:

1. Batched Requests

To streamline workflows, you’ll soon be able to batch up to 16 image generation requests into a single API call. This will not only provide results more quickly and reliably but also help you avoid rate limits, as the batch will count as one request.

2. Privacy-Enhanced Requests

We’re introducing an optional privacy mode for API calls. With this feature, you can receive results as a base64-encoded image or via a pre-signed URL with a short expiration time. No input data will be stored on our servers, ensuring compliance with regulations like GDPR. However, please note that this option will limit our ability to provide support and debugging assistance in case of issues.

There are additional features our users have requested that we agree are valuable and should be implemented. However, it will take us a bit longer to get to them:

Webhooks for Status Endpoint

Polling our status endpoint to retrieve try-on results doesn’t provide the best user experience. Once we’ve completed the migration of our API stack, we’ll revisit implementing webhooks as an alternative for our API users.

Automatic Top-Up

We understand the need for a feature that allows users to add payment details and automatically charge their card to top up credits when they’re running low. However, this requires careful research to ensure that automated payments are handled securely and responsibly, as there are inherent risks involved in such processes.

Infrastructure

Our software and hardware efforts are the backbone of our web app and API, ensuring a smooth, reliable, and consistent experience for all users. Here’s how we’re planning to meet growing demand:

1. API Stack Enhancements

Maintaining our API over the past few months has been a valuable learning experience. Based on these insights, we’re gradually shifting our stack to platforms that can process requests faster and with greater reliability.

2. Inference Partners

We’re in discussions with potential partners who specialize in hosting AI models. Partnering with a managed service provider will alleviate much of the operational responsibility from our small team, ensuring that our models are maintained by professionals with a focus on performance and uptime.

Closing Words

As we continue to grow, we remain committed to delivering innovative solutions that meet the needs of our users. The features and improvements outlined in this roadmap reflect our dedication to refining the FASHN platform, with your feedback playing a central role in guiding our progress. While some features are still in development or research, we’re excited about the future and look forward to sharing these updates with you as they come to life. Thank you for your ongoing support and for being part of this journey.