Image source: generated by Unbounded AI

In the past two years, large models, as a cutting-edge technology, are gradually penetrating into all aspects of the e-commerce industry.

In 2025, this change is still accelerating: Recently, the [Taobao Star · Tusheng Video] tool has been launched and is officially open to Taobao and Tmall merchants!

Experience address: https://agi.taobao.com/

(Currently only open to Taobao and Tmall merchants, ordinary users please wait patiently~)

1. Taobao Star: I understand you, and I understand e-commerce better!

[Taobao Star] is a video AIGC tool launched by Alimama based on the self-developed Taobao Star video generation model. Relying on advanced self-developed large models, massive e-commerce data, rich design language and industry-leading marketing experience, the [Taobao Star·Tusheng Video] tool demonstrates the following core advantages in e-commerce video generation:

Better understand product display methods: Based on rich e-commerce data and design experience, automatically identify and apply the most suitable display methods for different categories of products. From the precise presentation of product details to the unified coordination of the overall style, we ensure that every product can be displayed in the best possible way. Stronger multilingual semantic compliance capabilities: Deeply understand reference pictures and multilingual instructions (Chinese/English), accurately capture key action information in text instructions, and ensure that the generated video is highly consistent with user intentions. Reasonable physical and action laws: Strictly follow physical laws and action specifications to ensure smooth and natural images, avoid unrealistic image effects, and enhance the realism and professionalism of the video. Stable maintenance of characters, products and decorations: Whether it is faces, products, text, stickers and other decorative elements, they always remain clear and complete without jitter or deformation, improving visual consistency.

2. Taobao Star Video Generation Large Model

In the context of mainstream AIGC, video Generation is no longer a simple optimization and application of generation algorithms, but a more complex and systematic algorithm engineering, in which people, data, models, and computing power are the four core factors.

In March 2024, we developed and launched the Atongmu animation video based on the Unet Diffusion Model route, realizing the first application of video AIGC in the e-commerce scenario, and causing a certain spread in the industry degree and influence. However, due to the limitations of technical routes, data and other factors, Atongmu animation video still has a lot of room for improvement in key elements such as picture stability, content controllability, and richness of goals and actions.

Since March last year, we have been firmly and continuously investing in the direction of video AIGC. by algorithm, data, engineering and other teams, we have completed the data flywheel for generation tasks, self-developed 3D VAE, self-developed Taobao Star video generation large model (including Tbstar-T2V, Tbstar-I2V, Tbstar-V2V, etc.), etc. Construction of multiple key modules and models from 0 to 1. At present, the Tbstar-I2V large model has been used in the [Taobao Star · Tusheng Video] tool to provide users with the Tusheng Video function that understands e-commerce better.

Taobao Xingchen video generation large model

3. Tusheng videos that understand e-commerce better

3.1 Easy to operate

You only need to enter a static picture, Taobao Star can directly generate a video based on the understanding of the picture, or follow the text description entered by the user Generate videos, generate high-quality 5-second videos with one click.

[Taobao Stars·Tusheng Video] Tools

3.2 Better understanding of product display techniques

Display clothing and non-clothing products in a way that better understands e-commerce , especially the smoothness and professionalism of the model's movements are significantly improved.

prompt: A model shows clothes with a cool expression on her face

prompt: The model holds coffee, walks towards the camera, and shows clothes

prompt: levitates Bubbles and water droplets flutter

prompt: The product does not move, the plants sway with the wind, and the clouds move

3.3 Stronger multilingual semantic compliance capabilities

In-depth understanding of reference images, support for Chinese and English input, accurate capture of key action information in text instructions, ensuring that the generated video is highly consistent with user intentions.

prompt: The model covers her face with her mobile phone and takes a selfie in the mirror to show her clothes

3.5 Maintain stable characters, products and decorations

Especially for common text, stickers and other decorative elements in e-commerce images, they are always clear and complete without jitter or deformation, improving visual consistency.

prompt: a woman Model shows clothes to the camera

prompt: The camera retreats, the female model walks towards the camera and shows clothes

p>

Four. High-quality and low-cost native applications

Combined with explanation videos, template videos, etc. Post-editing tools enable the low-cost, intelligent production of high-quality, native e-commerce videos and apply them to mainstream e-commerce videos.business scene, and pan-entertainment scene.

Application 1: Product main image video supply

Generate main image video from a single product main image with one click.

Single main picture vs main picture video

Application 2: Selling point and eye-catching video

Generate video clips based on multiple model pictures, using template video method Make short product videos, add transitions, special effects and text to quickly generate eye-catching product display videos.

Application 3: Integrated clothing solution

“Model pictures” or “tile pictures” + virtual models can easily generate diversified clothing videos.

Different SKU tiles | Virtual model outfit pictures | Multi-SKU clothing videos

Application 4: Virtual try-on video

Based on user virtual try-on pictures Generate a virtual try-on video with one click to simulate the real upper body effect of clothing.

User virtual try-on pictures

Application 5: Video of UGC scenes

Convert UGC pictures into videos with one click to improve video supply, such as reviews, buyers Show, hang out, etc.

Application 6: Pan-entertainment scene

You can live in the AIGC era but you cannot live without work~

prompt: Superheroes and dinosaurs are walking slowly on the beach , with the city skyline in the background