Virtual Try-On and Product Holding Images: Bria and Kling
💡 One-sentence summary: Use ComfyUI to build a garment try-on image pipeline — Bria for try-on and outfit swap, Bria for product holding, Kling for high-resolution output.
What apparel products lack most isn't white-background images, but realistic shots of garments "worn on the body" or "held in hand." This lesson walks you through three official cloud templates to get ComfyUI virtual try-on and product holding images done.
Bria Virtual Try-On dresses your garment on a person photo, Bria Product Holding has the model pick up your product, and Kling Virtual Try-On produces high-resolution try-on images up to 2048px on the longest edge. All three are cloud API nodes: no local model downloads — upload images, spend credits, get output.
What you'll be able to do after this lesson
- Load the official "Bria: Virtual Try-On" template to "dress" a garment product image onto a person photo, keeping face, pose, and background intact
- Use the "Bria: Product Holding" template to have a model hold your product in a lifestyle image, with up to 3 multi-angle images of the same product
- Use the "Kling: Virtual Try On" template for high-resolution try-on images: output follows the person photo's dimensions, with the longest edge at 2048px
- Be able to explain how to choose among the three routes: Bria try-on and product holding at 12.07 credits each per run, Kling at 33.79 credits per run; shoes, bags, and accessories are not supported by Kling
Why use ComfyUI instead of web-based AI try-on tools?
| Capability | Web-based AI try-on tools | ComfyUI Cloud nodes |
|---|---|---|
| Billing | Mostly monthly/per-image packages | ✅ Pay-per-run credits, charged only for what you use |
| Image pipeline | Upload to web, wait for results, download manually | ✅ Node connections plug directly into cutout, upscaling, and saving |
| Product holding images | Dedicated categories are rare | ✅ Bria Product Holding specializes in "held in hand" |
| Swap multiple items at once | Usually one item at a time | ✅ Bria handles 1–3 items at once; two items are automatically composited into one layered outfit |
| Resolution | Depends on the site's output | ✅ Kling up to 2048px on the longest edge; Bria also offers 9 fixed aspect-ratio presets |
| Getting started cost | Just open a webpage and go | ❌ Requires installing ComfyUI, logging into an account, and having credits ready |
Your current pain points
- Apparel launches need worn-on-body images: hiring models, booking studio shoots, retouching — the whole process is slow and expensive
- Product detail pages need lifestyle context: you have white-background images, but "model carrying it out on the street" lifestyle shots are missing
- You've tried web-based try-on tools: uploading images one by one, waiting one by one, then manually downloading, cutting out, and upscaling the results — it never chains into a pipeline
When to use this technique
When you need to:
- Create AI outfit swap effect images: garments and accessories worn on the body, no models or studio needed
- Produce "model holding the product" lifestyle images — works for beauty, electronics, and beverages alike
- Quickly swap multiple outfits on the same person photo, or put on top and bottom pieces together in one go
And you don't want to:
- Download multi-GB local models — all three nodes in this lesson run in the cloud (for the local model route, see Models guide)
- Fiddle with samplers and denoise — just upload images and write one instruction to get output
🎒 Preparation before starting
- Complete the E-commerce Images in Practice guide, so you know each lesson starts from an official template
- ComfyUI updated to the latest version (step one of the official partner node tutorial; the desktop version follows stable releases)
- Log in to your Comfy account and have credits ready: all three nodes are cloud API nodes billed per run, with no local model downloads (same credit pool as the Bria cloud nodes in the Product Cutout and Background Replacement lesson)
- Know how to load a workflow from the template library (if unfamiliar, revisit Template Library guide)
- Prepare one person photo (for product holding, hands visible and empty works best); prepare product images per route: holding needs 1–3 multi-angle images of the same product, try-on needs 1–3 garment images you want to wear
- Have a budget in mind: Bria try-on and Bria product holding at 12.07 credits/run each, Kling try-on at 33.79 credits/run
Follow along with me
Step 1: Load the "Bria: Virtual Try-On" official template
Why: The official template already wires up the node chain—you just swap in your images.
Open Workflow → Browse Templates, search for Bria: Virtual Try-On and load it; you can also download the official template JSON directly (api_bria_virtual_try_on.json, in the Comfy-Org/workflow_templates repository on GitHub).
You should see: six nodes on the canvas—a note, a portrait LoadImage, a garment LoadImage, BriaVirtualTryOn, SaveImage, and ImageCompare. Default parameters of the core node: instruction is put the shirt on the person, aspect_ratio is auto, seed 42, moderation off.
Step 2: Upload portrait and garment images and run your first virtual try-on
Why: Bria try-on works by "replace + re-fit"—the garment in the garment image replaces the corresponding clothing on the portrait and re-fits the body; the face, pose, and background stay untouched.
Steps:
- Portrait
LoadImage: upload a portrait photo (official examplebria_tryon_man.png, downloadable from the template's input directory) - Garment
LoadImage: upload a garment image (official examplebria_tryon_shirt.png); flat-lay and ghost-mannequin shots work just as well—images with a model wearing the item are not required - Click Queue or press
Ctrl+Enter
You should see: an upper-body image of about 1 megapixel, saved with prefix api_bria_virtual_try_on; drag the slider in ImageCompare to compare before and after the replacement.
💡 This is Bria's unique value
Add one more via the + slot on the node—garment_image_2—and the two garments will be combined intoone layered outfit (with consistent lighting and fit)—top plus skirt in a single run, no need to run twice and stitch images together.
Step 3: Use "Bria: Product Holding" to produce product-in-hand images
Why: Categories beyond clothing (beauty, electronics, food) don't need "wearing"—they need "holding".
Search Bria: Product Holding in the template library and load it (template JSON: api_bria_product_holding.json). The structure is similar to the try-on template: two LoadImage → BriaProductHolding → SaveImage (this template doesn't include ImageCompare):
person_image: an empty-handed portrait (official examplebria_holding_person.png)product_image_1: product image (official examplebria_holding_product.png); up to 3 images via the+slot—use the same product's front, packaging, and another angleinstruction(optional): e.g.hold the bag in one hand,face the label to the camera
You should see: the model naturally holding the product, with the face, pose, and background unchanged, output at about 1 megapixel. Two details: if the original portrait already shows a similar object in hand, the product willreplace it rather than appear as an extra one beside it; transparent-background PNGs cut out in the earlierbackground removal lesson can also be uploaded, but transparent areas are treated as white.
Step 4: Switch to "Kling: Virtual Try On" for 2048px high-resolution try-on images
Why: After Bria's re-rendering the output is about 1 megapixel; when you need higher resolution for main product images or detail-page heroes, use Kling—its output matches the portrait image size, with the longest side capped at 2048px.
Search Kling: Virtual Try On in the template library and load it (template JSON: api_kling_vto.json). Five functional nodes:
LoadImage (portrait) ──→ KlingTryOnNode ──→ SaveImageAdvanced
LoadImage (garment) ──↗ └──→ ImageCompare (before/after comparison)- Portrait: upload a front-facing or three-quarter-view photo for the best results (official example
model_colorful_outfit.png) - Garment: upload a single item (official example
green_floral_dress.png): product shots, flat lays, mannequins, and model-worn images all work keep_poseis on by default: it preserves the original pose; turn it off and the pose adapts to showcase the garment- Click Queue or press
Ctrl+Enter
You should see: a try-on image at the same size as the portrait (longest side no more than 2048px); inputs exceeding 2048px are downscaled before running. Compare side by side with the original in ImageCompare.
Step 5: Pick a route by category and budget, then verify three things
Why: The three nodes have different capability boundaries—choosing the wrong route wastes credits.
| Bria Virtual Try-On | Bria Product Holding | Kling Virtual Try-On | |
|---|---|---|---|
| What it does | Puts clothing/accessories on the body | Puts the product in hand | Puts clothing on the body (high-res) |
| Input | Portrait + 1–3 garments | Portrait + 1–3 images of the same product | Portrait + 1 garment |
| Output | About 1 megapixel | About 1 megapixel | Portrait size, longest side ≤2048px |
| Category boundaries | Clothing and accessories | 1–3 images of the same product | Clothing only: shoes, bags, and accessories not supported |
| Price | 12.07 credits/run | 12.07 credits/run | 33.79 credits/run |
| Additional controls | instruction / aspect_ratio / seed / moderation | Same as left | keep_pose / seed (results still non-deterministic) |
You should see: all three verification items check out—the face is unchanged, the pose is right, and the product details match the original image (logo, labels, colors). Kling may produce different results across two runs with the same settings; for important images, run several times and pick the best.
📋 Magic Prompt
The instruction field on Bria's two nodes is where instructions go. The official recommendation is to run once with it empty first, and only write something if you're not satisfied; the official examples are all in English, so writing in English is the safest.
📝 Holding-pose instructions (Product Holding)
Expected effect: control "how to hold it, which way the label faces".
hold the 【商品名,如 bag】 in one hand, face the label to the camera📝 Try-on detail instructions (Virtual Try-On)
Expected effect: leave the garment itself untouched and only adjust wearing details.
roll the sleeves upor
keep the shirt untuckedCheckpoint ✅
- All three templates are findable in the template library:
Bria: Virtual Try-On,Bria: Product Holding,Kling: Virtual Try On(if not found, update ComfyUI first) - Bria try-on runs successfully: face, pose, and background unchanged, garment fits the body, saved with prefix
api_bria_virtual_try_on - Two garments run in one pass into a single layered outfit
- Kling try-on run end to end: output matches the portrait's dimensions, with the longest side not exceeding 2048px
- Model selection is clear-cut: worn clothing / hand-held products go with Bria (12.07 credits each), high-definition clothing goes with Kling (33.79 credits), while shoes, bags, and accessories do not go through Kling
Pitfall Reminders
❌ Running Product Holding when the portrait's hand is clenched or holding something
Official tip: portraits with visible and empty hands work best; a clenched or occupied hand usually produces awkward results. Swap in a photo with empty hands first.
❌ Stuffing two different products into product_images
product_images accepts 1–3 images of the same product (front view, packaging, another angle). Mix in two different products and the rendering falls apart.
❌ Using a full outfit photo as the garment input
An image of a whole outfit replaces more than you expect—use single-item photos (on hanger or flat lay both work). If you want a full outfit, split it into two pieces and put them into garment_image_1 and garment_image_2, letting Bria composite a layered outfit.
❌ Running shoes, bags, or accessories through Kling try-on
Kling officially states clothing only: shoes, bags, and accessories are not supported. Route these products through Bria Product Holding's hand-held path instead.
❌ Expecting seed to lock in Kling results
Official note: no matter what the seed is, Kling try-on results are non-deterministic—two runs with the same settings may differ. Don't treat a fixed seed as a way to "reproduce" results.
Advanced Tips (Optional)
- Generating images with a different aspect ratio:
aspect_ratiodefaults toauto(preserving the portrait framing); there are 9 fixed aspect ratio presets in total—1:12:33:23:44:34:55:49:1616:9—outputting 0.6–1 megapixels (e.g., 3:4 is 768×1024). Use it when different platforms need different aspect ratios - Getting variants by changing Bria's seed: with the same input, vary
seedto explore different versions and pick the one that looks most like a "real photo" - The moderation switch: off by default; turning it on enables visual output moderation
- Chaining into the upscaling workflow: Bria outputs about 1 megapixel; for larger sizes, connect the upscaling workflow from the previous lesson—try-on images include backgrounds, so both the conservative upscale and the 2K diffusion routes work
- Chaining into the cutout workflow: to replace the output's background, go back to the Product Cutout and Background Replacement lesson to handle it
Lesson Summary
With the three cloud templates done, ComfyUI virtual try-on and hand-held product images are, for you, just a matter of "pick a node, upload images, write one instruction." ✅ You can now:
- Put clothing on a portrait with Bria Virtual Try-On, up to 3 items at a time, with two items composited into one outfit
- Use Bria Product Holding to have a model hold a product in a scene image
- Produce high-definition try-on images up to 2048px on the longest side with Kling Virtual Try-On
- Choose a route based on product category and budget, avoiding pitfalls like shoes/bags/accessories and mixed products
Key parameters at a glance:
| Parameter | Template Default | Description |
|---|---|---|
| instruction | try-on put the shirt on the person; hand-held hold the bag in one hand | can be left empty for a first run; official examples are in English |
| aspect_ratio | auto | Preserves portrait framing; the 9 fixed presets output 0.6–1 megapixels |
| seed | 42 (fixed, Bria template) | Change the seed in Bria to get variants; the Kling template uses a random seed with non-deterministic results |
| moderation | off | Switch for visual output moderation |
| keep_pose (Kling) | on | When off, the pose adapts to showcase the clothing |
Next Lesson Preview
Now that people can wear and hold things, the next lesson makes the product itself "stick" onto physical objects: Product Mockups: Applying Designs to Physical Objects. You will learn:
- Using official templates to apply designs and patterns to physical objects to produce mockup images
- How to choose between FLUX.2 mockup workflows and the product placement template
- Image-generation techniques for "object-on-object" scenarios like packaging and vehicle bodies