Overview: Why OpenAI’s New Shopping Tools Matter
On October 1, 2026, OpenAI announced the global rollout of two shopping‑centric features inside ChatGPT: a Virtual Try‑On experience and a Favorites library. Both are powered by the freshly released ChatGPT Images 2.5 model, which OpenAI says delivers “more natural lighting and richer textures, follows editing instructions more reliably, and reduces image generation latency.”
The significance of these additions goes beyond a simple UI tweak. They embed generative‑AI directly into the purchase funnel, turning a text‑only chatbot into a visual shopping assistant that can:
- Render realistic clothing previews on a user’s body in seconds.
- Persist product selections alongside generated try‑on images for later reference.
- Bridge the gap between inspiration (Pinterest‑style discovery) and conversion (checkout).
In a market where Google, Pinterest, and emerging agentic‑AI startups like Instinct are already experimenting with visual commerce, OpenAI’s move signals that large‑language‑model platforms are now serious contenders in the e‑commerce ecosystem.
Technical Breakdown: The Images 2.5 Engine
Core Improvements Over Previous Models
The Images 2.5 model is a diffusion‑based generator tuned specifically for fashion‑related prompts. Its enhancements include:
🔹 ---------
• Benefit: ---------
🔹 *Natural lighting*
• Benefit: Reduces the “studio‑look” artifact, making garments appear as they would in everyday environments.
🔹 *Richer textures*
• Benefit: Captures fabric nuances—silk sheen, denim weave, leather grain—critical for accurate fit perception.
🔹 *Instruction fidelity*
• Benefit: Better adherence to user edits such as “make the shirt more fitted” or “show the dress in daylight.”
🔹 *Latency reduction*
• Benefit: Average generation time drops from ~3 seconds to under 1.2 seconds on typical consumer hardware.
These gains are the result of a larger training corpus that mixes high‑resolution runway photography with user‑generated selfies, plus a new conditioning pipeline that aligns pose estimation with garment segmentation
…that aligns pose estimation with garment segmentation, allowing the model to accurately drape clothing over a user’s body shape and posture. OpenAI also introduced a lightweight on‑device inference cache that stores recent pose embeddings, shaving milliseconds off each subsequent try‑on request.
Prompt Engineering for Fashion
OpenAI exposed a set of “style tokens” that developers can embed in prompts to steer the visual output:
-
--fabric silk– emphasizes sheen and smoothness. -
--lighting sunset– simulates golden‑hour ambience. -
--fit relaxed– widens the silhouette for a looser look.
These tokens are optional for end‑users but power the underlying system to translate natural‑language tweaks (“make the jacket more fitted”) into concrete image‑generation parameters.
How the Virtual Try‑On Feature Works
- Upload a Reference Image – Users can drop a selfie, a full‑body photo, or even a screenshot of a product page. The app runs a face‑and‑body detector to extract key landmarks.
- Select a Product – ChatGPT surfaces a carousel of items matching the user’s query (e.g., “summer dresses under $150”). Each tile includes a “Try On” button.
- Real‑Time Rendering – When pressed, the Images 2.5 engine composites the selected garment onto the user’s pose, applying the appropriate lighting and fabric texture. The result appears within 1.2 seconds on average.
- Iterative Edits – Users can ask follow‑up questions like “show me the dress with a belt” or “swap the color to navy,” and the model regenerates the image while preserving the original body pose.
- Save or Share – The final render can be saved to the new Favorites library or shared directly to messaging apps, email, or social platforms.
Edge Cases Handled
- Partial Body Shots – If only a torso is visible, the system extrapolates the missing limbs using a generative pose estimator, ensuring the garment fits plausibly.
- Multiple Users – In group photos, the model can isolate each person and apply different outfits simultaneously, a feature useful for family shopping trips.
- Accessibility – For visually impaired users, ChatGPT can describe the generated look aloud, citing fabric feel, cut, and color contrast.
The Favorites Library: A Persistent Shopping Hub
The Favorites tab lives inside the ChatGPT sidebar under “Shopping.” When a user clicks “Save to Favorites,” the following occurs:
🔹 --------
• What Happens: --------------
🔹 *Product Metadata Capture*
• What Happens: SKU, price, retailer link, and any discount codes are stored alongside the generated image.
🔹 *Versioned Try‑On Snapshots*
• What Happens: Each time a user revisits a product and tweaks the look, a new snapshot is appended, creating a visual history.
🔹 *Cross‑Device Sync*
• What Happens: Favorites sync via the user’s OpenAI account, making them accessible on mobile, desktop, or the upcoming ChatGPT‑plus smartwatch app.
🔹 *Export Options*
• What Happens: Users can export a PDF lookbook, download a CSV of product URLs, or push the list to a connected e‑commerce cart (e.g., Shopify, Amazon).
The library also integrates with OpenAI’s Shopping Assistant AI, which can proactively suggest complementary items based on the saved wardrobe, nudging users toward complete outfits.
Competitive Landscape: Who’s Already Doing This?
🔹 ---------
• Feature: ---------
• Launch Year: -------------
• Notable Edge: --------------
🔹 *Google*
• Feature: “Lens Try‑On” (AR overlay)
• Launch Year: 2023
• Notable Edge: Real‑time AR on Android devices, no server‑side rendering.
🔹 *Pinterest*
• Feature: “Shop the Look” visual search
• Launch Year: 2024
• Notable Edge: Community‑curated style boards, strong social discovery.
🔹 *Instinct*
• Feature: “Proactive Recommendations”
• Launch Year: 2025
• Notable Edge: Agentic AI that pushes product alerts based on user behavior.
🔹 *OpenAI*
• Feature: Virtual Try‑On + Favorites
• Launch Year: 2026
• Notable Edge: Diffusion‑based photorealism, seamless text‑to‑image loop, integrated LLM context.
While Google’s AR solution excels on-device, it lacks the generative flexibility to change garment attributes (e.g., color, fit) on the fly. Pinterest offers inspiration but does not render the user wearing the items. Instinct’s proactive nudges are powerful but still rely on static product images. OpenAI’s approach uniquely blends generative visual synthesis with conversational context, allowing a back‑and‑forth dialogue that feels more like a personal stylist than a static catalog.
Pricing, Availability, and Rollout
- General Availability: The features are live globally for all ChatGPT users on the free tier, with no additional charge.
- Premium Enhancements: ChatGPT Plus subscribers receive higher‑resolution renders (up to 4K) and priority access to the “Batch Try‑On” mode, which can process up to 10 items in a single request.
Read the full breakdown originally published at https://ltdeveloperblogs.github.io/posts/chatgpt-can-now-virtually-try-on-clothes-for-you/
Top comments (0)