-
-
Same pose. Same outfit. Better expression. BESTFIT combines YouCam AI Clothes and AI Retake into one shopping-photo workflow.
-
BESTFIT turns separate visual assets into one more useful virtual try-on experience.
-
Three inputs, one better try-on: the pose you want to keep, the garment you want to try, and the expression you prefer.
-
Real YouCam execution proof with Clothes, Retake Pre-process and Retake task IDs plus input SHA-256 hashes.
Inspiration
Virtual try-on usually starts from a single source photo.
But one photo rarely contains everything a shopper wants to keep.
You may have one image with the right pose and framing, but a neutral or awkward facial expression. Another photo may have the smile or expression you actually like, but not the right composition for a fashion try-on.
That led to a simple question:
Why should your best virtual try-on have to come from one photo?
BESTFIT explores a different model: use the best photo for each part of the experience.
- one photo for the pose,
- one garment to try on,
- one photo for the preferred expression.
Then combine them into one better shopping image.
The core idea became:
Wear the outfit. Keep your best face.
What it does
BESTFIT takes three inputs:
- Best pose photo — the body pose and framing the shopper wants to keep.
- Garment — the fashion item they want to try.
- Best expression photo — another photo of the same shopper with the facial expression they prefer.
BESTFIT first sends the pose photo and garment to YouCam AI Clothes V4.
That produces the outfit try-on while keeping the original pose and composition.
The generated result is then processed by YouCam AI Retake together with the shopper's preferred-expression photo.
The final output preserves the useful pose and outfit while restoring the expression the shopper actually likes.
The result is:
Same pose. Same outfit. Better expression.
The prototype also includes a Judge Demo that loads a validated example and walks through the entire experience in a few clicks.
A technical Proof Receipt records the real YouCam task IDs, detected-face count, and SHA-256 hashes of the input assets.
How we built it
BESTFIT is a lightweight Node.js web application built directly around the YouCam REST APIs.
The workflow is:
Best pose + Garment
↓
YouCam File API
↓
YouCam AI Clothes V4
↓
Outfit try-on
↓
Upload generated result
↓
YouCam Retake Pre-process
↓
Validate one usable source face
↓
Favorite expression photo
↓
YouCam AI Retake
↓
BESTFIT final
The main YouCam capabilities used are:
- YouCam File API
- YouCam AI Clothes V4
- YouCam AI Retake
- YouCam Retake Pre-process
The application uses asynchronous task creation and polling for both AI Clothes and AI Retake.
The deployed prototype uses a stateless serverless workflow: provider state is carried securely between the Clothes and Retake stages, while the browser receives only the data required to continue the validated pipeline.
The UI was intentionally kept simple and consumer-facing:
- upload the three inputs,
- click Try the outfit,
- click Keep my best face,
- compare Original → Outfit Try-on → BESTFIT Final.
No LLM, recommendation engine, login system, or unnecessary backend infrastructure is used in the core flow.
Challenges we ran into
The biggest technical challenge was integrating YouCam AI Retake.
The live API behavior differed from the first payload patterns we implemented from the documentation examples.
The first pre-process request failed because the live endpoint expected a flat:
src_file_id
rather than the nested payload shape we initially used.
We then discovered that the task status was retrieved using the task ID in the URL path rather than the query parameter we first tried.
For the final Retake request, the live validation responses revealed the accepted fields:
src_file_url
ref_file_ids
face_mapping
We solved this by breaking the integration into independently testable stages:
- file upload,
- Clothes generation,
- Clothes result re-upload,
- Retake pre-processing,
- face detection validation,
- reference expression upload,
- final Retake.
This made it possible to inspect the real API responses at every step and avoid guessing.
Another challenge was designing the experience so it felt like a product rather than a sequence of API calls.
We solved that by building the entire UX around one visible transformation:
Original → Outfit Try-on → BESTFIT Final
Accomplishments that we're proud of
We are especially proud that BESTFIT went beyond a simple single-endpoint virtual try-on.
The prototype successfully combines two distinct YouCam capabilities into one coherent consumer experience.
The complete end-to-end flow is working:
- AI Clothes generation succeeds,
- Retake pre-processing detects the source face,
- AI Retake succeeds,
- the final expression visibly changes,
- the outfit and pose remain usable.
We also built a reproducible Judge Mode and a technical Proof Receipt using real provider task IDs and input hashes.
Most importantly, the final result communicates the product idea visually without requiring a long explanation.
What we learned
The biggest product lesson was that virtual try-on does not have to treat the user's source photo as an indivisible object.
Different photos can contain different useful components.
One photo can provide the pose.
Another can provide the preferred facial expression.
A product image provides the garment.
YouCam's APIs make it possible to compose those elements into a new commerce workflow.
We also learned that the strongest hackathon idea was not necessarily the most complex one.
BESTFIT became much stronger when we reduced the product to one consumer problem and one obvious visual outcome.
What's next for BESTFIT
The next version of BESTFIT could extend the same idea into a broader commerce-photo composition layer.
Potential directions include:
- fashion retailer integrations,
- creator and affiliate shopping content,
- marketplace seller imagery,
- fashion rental platforms,
- social-commerce previews,
- saved BESTFIT looks,
- direct product-catalog integration,
- multiple outfit variations,
- mobile camera capture guidance.
A future production version could also automatically validate whether the pose and expression references belong to the same user before generation.
The long-term idea is broader than one try-on:
your best commerce image should be able to use the best parts of the visual assets you already have.
Built With
- computer-vision
- css3
- ecommerce
- fashion-tech
- html5
- javascript
- node.js
- rest-api
- virtual-try-on
- youcam-ai-clothes
- youcam-ai-retake
- youcam-api
- youcam-file-api