Business

Understanding Seedance 2.0’s Multi-Modal Input: My First Project

Published

3 weeks ago

21 February 2026

Understanding Seedance 2.0's Multi-Modal Input: My First Project

When I first heard about “multi-modal input,” it sounded intimidating. Images, videos, audio, text—all working together in a single video generation? I wasn’t sure how that actually worked in practice, or if I even needed all those features.

But once I started experimenting with Seedance 2.0, I realized the multi-modal capability wasn’t a complicated luxury feature; it was actually the simplest way to create better videos.

Let me walk you through my first real project using multi-modal input, and what I learned along the way.

What I Thought Multi-Modal Input Would Be

Before I actually tried it, I had some misconceptions. I imagined it would require technical skill—like some sort of advanced prompt engineering where I’d need to specify exactly how each file interacted with every other file. I thought I’d need to understand the “rules” of combining images with audio, or know the exact syntax for referencing multiple inputs.

The reality was much simpler.

Multi-modal input just means you can throw different types of files at Seedance 2.0 and tell the model what you want it to do with them. That’s it. You’re not switching between different tools or learning a special command language. You’re just giving the model more information to work with.

My First Project: A Short Brand Story Video

I was approached by a local coffee roastery that wanted a 10-second promotional video. They had given me:

Three high-quality product photographs of their different bean varieties
A 5-second video clip of someone pouring coffee into a cup (they’d shot it themselves)
A 3-second audio clip of coffee brewing sounds
A brief description of the mood they wanted: “warm, inviting, craft-focused”

Normally, I would have had to choose between using the images OR the video OR the audio in post-production. I’d create one asset and try to make it work, leaving other materials unused.

With Seedance 2.0’s multi-modal capability, I could use everything at once.

How I Actually Set It Up

Step One: Gathering the Assets

The coffee roastery gave me three product photos, a pouring video, and brewing sound effects. I organized these before uploading, though honestly, I could have just uploaded them randomly—the point is that Seedance 2.0 can handle all of it simultaneously.

Step Two: Uploading Everything

Seedance 2.0 lets you upload:

Up to 9 images
Up to 3 videos (total duration ≤15 seconds)
Up to 3 audio files (total duration ≤15 seconds)
Text descriptions of unlimited length

For my project, I uploaded all three product photos, the pouring video, and the brewing audio. The platform accepted everything without complaint.

Step Three: Writing a Natural Language Description

This was the key part that surprised me. I didn’t need to learn special syntax. I just described what I wanted, referencing the files by number or type.

My prompt looked something like this:

“Create a 10-second promotional video. Start with a close-up of @image1 (the espresso beans), with the coffee brewing sounds from @audio1 playing underneath. Transition smoothly to @video1 (the pouring shot), with the warm, crafted aesthetic of @image2 visible in the background. End with a final shot of @image3 (the roasted beans close-up) with the brewing sounds fading out. The overall mood should be warm and inviting, like a specialty coffee shop experience.”

That was it. Natural language. No special operators or complex syntax.

What Happened When I Generated

I honestly wasn’t sure what to expect. Would it use all the files? Would it ignore some of them? Would it misunderstand my descriptions?

The first generation was surprisingly good. The video opened with the espresso beans from my first image, the audio played throughout, and the pouring shot appeared in the middle. The transition between the still image and the video felt natural, not jarring. The final product felt cohesive in a way that would have been really difficult to achieve with traditional video editing.

Was it perfect? No. There were a few things I’d adjust on the second try. But the point is that all my different media assets—photos, video, and audio—came together into a single coherent video without me having to manually edit them together.

Why This Matters for My Workflow

Before understanding multi-modal input, I was used to this process:

Choose one primary asset (usually video or images)
Create supplementary graphics or transitions in editing software
Add audio in post
Export the final video

It was time-consuming and resulted in a patchwork feel—pieces assembled together rather than something that felt naturally integrated.

With multi-modal input:

Gather all assets (images, video, audio, description)
Upload everything to Seedance 2.0
Describe what I want
Get a generated video with all elements incorporated
Make minor tweaks if needed

The second workflow is faster and produces more cohesive results because the model synthesizes everything together from the start, rather than me trying to glue separate pieces together afterward.

Real-World Examples of Multi-Modal Combinations

Since that first project, I’ve experimented with different combinations:

Education Videos

I’ve used reference images of diagrams, a short video clip showing a concept in action, and a voiceover audio track explaining what’s happening. The model generates a video that incorporates the visual information, the dynamic demonstration, and the audio explanation all at once. Students get a more complete learning experience than if I’d just picked one format.

E-Commerce Product Demonstrations

Multiple product photos + a video showing the product in use + background music = a more engaging product video than I could create with any single asset type alone. The images establish what the product looks like, the video shows it functioning, and the audio creates the right emotional tone.

Social Media Clips

For Instagram Reels, I’ve combined a still image of the caption text I want to appear, a short video of motion that fits the content, and upbeat audio. The multi-modal approach ensures all elements appear in the final video without me manually compositing them.

The Learning Curve

Honestly, there wasn’t much of one. The main thing I had to learn was to be more specific about which asset I wanted referenced where. In my first few attempts, I was vague—like, “use the images throughout the video”—and the results were less predictable.

Once I started being explicit—”start with image1, transition to video1, end with image3″—the model understood my intent better. The specificity improved the results significantly.

The other lesson was that quality varies across asset types. My higher-resolution images worked better than low-res ones. My stable video clips worked better than shaky handheld footage. This isn’t surprising, but it’s worth noting: garbage input still produces less impressive output, even with AI.

Limitations I’ve Hit

Multi-modal input is powerful, but it has boundaries. If I upload too many assets and ask the model to incorporate all of them in a short 5-second video, the result feels rushed or cluttered. There’s a reasonable ratio of content to output duration.

Additionally, if the audio I provide has specific timing—like a voiceover with precise pauses—the model doesn’t always match the visual content to those exact timestamps. It’s close, but not frame-perfect. For critical applications like lip-sync, I might need to make adjustments afterward.

Complex interactions between assets can also be unpredictable. If I upload a video where the person is wearing a blue shirt and a photo where they’re wearing red, the model might struggle with consistency. It works better when reference materials are conceptually compatible.

Why I’m Now a Multi-Modal Believer

The practical benefit is this: I can incorporate more creative assets into my videos without doing manual video editing. That means faster turnaround times and more polished final products. It means I can use all the reference material a client gives me, rather than having to choose which piece to prioritize.

For freelancers and small teams, that’s genuinely valuable. It removes a technical bottleneck from the production process.

Moving Forward

I’m still exploring what multi-modal input makes possible. I’ve started experimenting with edge cases—like uploading multiple audio tracks to see how the model combines them, or using reference images and videos that have very different aesthetics to see if the model can synthesize them into something cohesive.

The feature isn’t a magic fix for poor planning or low-quality assets. But if you gather good reference material and think clearly about what you want to create, Seedance 2.0‘s multi-modal capability can genuinely simplify your creative process.

For anyone who’s used to assembling videos from different pieces in post-production, this approach feels like a meaningful step forward. You’re describing your vision once, clearly, and the model generates something that incorporates all your reference materials from the start. That’s the real power of multi-modal input.

Business

Sadanand Date takes charge as Sebi executive director

Published

7 minutes ago

14 March 2026

NewsAdmin

Sadanand Date takes charge as Sebi executive director

Sadanand Date assumed charge as Executive Director at Sebi on March 4 to head the investigations department, the markets regulator said on Friday.

Date is a 2007-batch IPS officer of the Uttarakhand cadre.

Prior to joining Sebi, he was on central deputation to the Central Bureau of Investigation (CBI), where he served in several key roles, including Superintendent of Police in the Anti-Corruption Branch (ACB) and Bank Securities and Fraud Cell (BSFC), the regulator said in a statement.

He also headed multiple branches in Mumbai, including the Economic Offences Branch, Special Crime Branch, Special Task Branch and Anti-Corruption Branch.

During his tenure with Uttarakhand Police, Date held several leadership positions and served as Superintendent of Police or Senior Superintendent of Police in various districts, such as Uttarkashi, Nainital, Haridwar, Udham Singh Nagar and Dehradun.

BSE, NSE organise mock trading session today: Check timing, purpose, other details

Stock exchanges BSE and NSE are conducting mock trading sessions for equity, commodity and currency derivatives on Saturday from the primary site (PR) and Disaster Recovery Site (DR). The mock trading is merely for the purpose of testing and familiarisation. The trades resulting from such mock trading will not attract any margin obligation or pay-in and pay-out obligation, and they will not create any rights and liabilities.

Trading members using third-party trading platforms can also use this opportunity to test their respective trading applications during the mock trading session for various functionalities (including exceptional market conditions), viz., various types of call auction sessions, risk-reduction mode, trading halt, block deals, etc.

Here’s the schedule of trading sessions:

– Log-in – 09:15 am to 09:45 am
– Morning Block Deal Window (PR): 09:45 am to 10:00 am
– Continuous Trading T+1 (PR): 10:15 am to 01:00 pm
– Continuous Trading T+0 (PR): 10:15 am to 12:30 pm

– Closing: 04:00 pm to 04:10 pm
– Post-closing: 04:10 pm to 04:20 pm
– Trade Modification T+1: 04:30 pm
– Trade Modification T+0: 03:45 pmThe exchanges have urged market participants to participate actively in the mock trading sessions.

Exchanges routinely conduct mock trading sessions to test their systems to be able to provide their members with a robust & efficient system for trading with better features.

They also seek feedback from all members. The members can give their feedback for the mock trading session to exchanges by 5:00 pm.

Indian benchmark indices fell sharply on Friday, recording their third successive decline as the Iran-Israel/US war continued to dent market sentiments. The biggest drags were metals, auto, and financial stocks. In a volatile session, the broader Nifty plunged 488.05 points, or 2.06%, to close at 23,151.10, while the 30-share Sensex declined 1470.50 points, or 1.93%, to settle at 74,563.92.

(Disclaimer: The recommendations, suggestions, views, and opinions given by the experts are their own. These do not represent the views of The Economic Times.)

Business

How systematic active investing combines data, discipline and dynamic allocation to help deliver alpha

Published

57 minutes ago

14 March 2026

NewsAdmin

How systematic active investing combines data, discipline and dynamic allocation to help deliver alpha

Investing has traditionally been shaped by human judgment—port analysing companies, interpreting macroeconomic signals, and making decisions based on experience and intuition. While this approach has produced many successful strategies, it is inherently constrained by human bandwidth and individual bias.

Considering the above, research teams can track a limited number of companies, process a finite volume of information, and react within time-bound constraints.

Systematic investing represents a meaningful evolution in this framework. It combines human expertise with machine-driven analytical power to create a more structured and scalable investment process.

In essence, systematic investing brings together two complementary strengths:

Human insight — experience, judgment, and economic understanding
Machine intelligence — speed, scale, and analytical precision

This fusion allows the investment team to analyse vast datasets, evaluate market signals in real time, and apply consistent decision-making frameworks.

The result is an investment approach that is disciplined, repeatable, and resilient, which are qualities that are increasingly valuable in modern markets.

Why India Is an Ideal Market for Systematic Investing

India’s capital markets are undergoing a structural transformation. Over the past decade, the ecosystem has been shaped by several powerful trends, these include rapid growth in retail investor participation, Digitisation and faster dissemination of information, increasing market depth and sectoral diversity along with Greater liquidity and trading activityIn such an environment, the ability to process information quickly and identify signals efficiently can become a powerful competitive advantage.

This is where the Systematic Active Equity (SAE) strategies stand out.

SAE combines the alpha-seeking intent of active management with rules-based, data-driven execution frameworks that are cost-controlled and risk-managed. This allows investment strategies to identify opportunities more efficiently and implement them with discipline and precision at lower cost.

The Core Pillars of Systematic Active Equity

1. Data-Driven Decision Making at Scale

One of the defining characteristics of SAE strategies is their ability to process vast and diverse datasets. These include traditional financial metrics such as earnings, valuations, balance sheet indicators, Market-based signals like price momentum and liquidity trends. Furthermore, the strategies also include Alternative datasets such as News sentiment analysis, Social media signals, Satellite and geospatial data, amongst others.

The objective is to identify repeatable patterns and predictive signals that can inform investment decisions. Over time, models continuously learn from new information, refine their insights, and adapt to evolving market dynamics.

2. Dynamic and Adaptive Portfolio Construction

Unlike static portfolios or purely benchmark-hugging strategies, SAE portfolios are inherently dynamic. They continuously adjust based on:

Signal strength
Changing market conditions
Factor performance cycles

This enables portfolios to rebalance efficiently and allocate capital where opportunities looks strong. In markets like India—where sector leadership and market themes can rotate rapidly—this adaptability becomes an important source of investment edge.

3. Integrated Risk Management

Risk management in systematic strategies like SAE is not a separate layer applied after portfolio construction. Instead, it is embedded within the investment framework itself.

This includes:

Volatility targeting
Position sizing/weighting frameworks
Diversification across sectors and market caps
Active Risk control mechanisms
Analyzing factor exposures and tilting them based on strategy goals
Focusing on risk-return metrics like IR (Information Ratio) Alpha consistency as a target
Eliminating key-man risk

The goal is not only to generate returns but also to ensure consistency of outcomes across market cycles.

How Systematic Investing Reduces Behavioural Biases

Traditional discretionary investing, while driven by expertise, can sometimes be influenced by behavioural biases such as, Recency bias, Overconfidence etc

By reducing the influence of emotion and subjectivity, systematic strategies enable a more consistent and forward-looking investment process, thereby eliminating human biases by relying on , pre-defined investment rules, Data-backed signals and Objective decision frameworks

Ensuring Continuity Beyond Individuals

Another structural advantage of SAE lies in its process-driven nature. In traditional setups, fund performance can sometimes be closely associated with individual portfolio managers and hence lead to key man risk. Changes in personnel may lead to shifts in strategy or portfolio construction leading to very different risk and return orientations than originally anticipated. Systematic investing reduces this dependency. Despite changes in the investment team, the underlying models remain constant as data pipelines continue operating ensuring the overall investment philosophy remains undisturbed.

In many ways, it is like changing the driver while the navigation system guiding the journey remains the same.

Combining Human Expertise with Machine Precision

Despite common perception, systematic investing is not about replacing human decision-making. Instead, it is about augmenting human expertise with technology.

Humans play a critical role in

Designing robust and efficient investment frameworks is important to avoid GIGO (Garbage-In, Garbage-Out)

Selecting relevant signals
Interpreting macroeconomic context to decide on active risk levels
Monitoring and refining models

Machines, in turn, excel at:

Processing vast datasets
Identifying patterns across markets
Executing strategies with speed and consistency

Together, this partnership creates a powerful investment engine—where humans define the “what” and “why,” and machines optimise the “how” and “when.”

A New Paradigm for India’s Investors

As India’s markets become more complex, information-rich, and competitive, investors increasingly require strategies that can combine discipline, scalability, and adaptability.

Systematic Active Equity addresses this need by integrating:

Data-driven intelligence
Machine efficiency
AI/ML techniques
Human oversight and governance

The outcome is a robust and repeatable investment approach designed to navigate volatility, capture opportunities, and deliver alpha over time with controlled risk and reduced cost.

For Indian investors, this represents a shift towards a more institutional-grade investment framework incorporating global best practices.

(The author is CIO at JioBlackRock Asset Management)

Business

As crude oil price breaches $100 mark, Systematix recommends RIL, a potential multibagger and 4 more stocks to buy – Ripple Effect

Published

1 hour ago

14 March 2026

NewsAdmin

As crude oil price breaches $100 mark, Systematix recommends RIL, a potential multibagger and 4 more stocks to buy - Ripple Effect

The Iran-Israel war has entered its 15th day, causing crude oil prices to soar to $103 a barrel. They have increased by over 35% so far this year, and expectations are that they could hit the $150 mark if the war continues. In light of the ongoing crisis, brokerage Systematix Institutional Equities has recommended 6 stocks with a potential upside of 103%. Destruction of oil & gas assets amid the West Asia War triggered a strong risk premium in prices. Tightening supply dynamics—owing to the closure of the Strait of Hormuz, elevated tanker freight rates and insurance premiums for vessels—will keep prices high, helping upstream companies in its view.