DEV Community: Pranav Shah

I Used to Think AI Videos Were Slop. Then I Changed the Workflow.

Pranav Shah — Sun, 01 Nov 2026 11:34:26 +0000

I Used to Think AI Videos Were Slop. Then I Changed the Workflow.
For a while, whenever I saw an AI-generated video, my first thought was: “This is AI slop.”

The visuals might be impressive. The video might be smooth. But the message often felt empty—generic script, random visuals, a voice that could belong to anyone, and no real reason for me to keep watching.

I don’t think the problem was simply that AI made the video. The problem was asking AI to make content before deciding what I actually wanted to say.

That changed how I think about using AI for social media. I stopped looking for one magic “make and post everything” button and started building a repeatable workflow around an idea, a voice, and a little human judgment.

The tools in my current workflow
Claude Code + my custom skills: the brain. I use Claude Code to coordinate the work. My custom skills give it repeatable instructions—what I’m trying to communicate, how I want the content to sound, and what needs to be prepared for each channel. A skill is useful because I don’t have to re-explain the whole process every time; Anthropic documents skills as instructions kept in SKILL.md files and loaded when relevant.

Higgsfield: visual material. I use Higgsfield as part of making the video itself—creating or animating visuals and directing the kind of shot I want. Its image-to-video workflow, for example, can animate a still and let you guide camera movement.

HyperFrames: putting the video together. HyperFrames lets an AI coding agent build a video composition using web technologies and render it as a video. I use it for things like arranging scenes, captions, titles, and motion into a more consistent finished piece. It’s open source and built to work with coding agents.
**
Ootto**: Instagram. In my current setup, Ootto is the tool I use to manage Instagram. I’m not presenting it as a one-click publisher for every social network; it’s the Instagram part of my workflow.

For me, the useful change is that Claude Code can act as the coordinator, the video tools help with production, and Ootto handles the Instagram side. The goal isn’t to make every platform sound the same. It’s to start from one real idea and shape it for the place where people will see it.

One idea, five different posts
Let’s use the idea behind this article: AI videos used to feel like slop to me, and a more thoughtful workflow changed my mind.

Here’s how I can turn that one idea into useful content without copy-pasting the same caption everywhere:

Start with the point. I write down what I actually believe: AI can speed up production, but it can’t rescue an idea that has nothing to say. I also note who the post is for and what I don’t want to claim.

Ask my Claude Code skill for a content pack. It can help me shape the hook, a short script, and separate drafts for each channel. I can also ask it to flag statements that need fact-checking and avoid inventing numbers or personal experiences.

Make visuals that serve the message. I use Higgsfield to create or animate supporting shots. Instead of asking for “something cinematic,” I can give it a subject, an action, a mood, and a camera direction.

Assemble the video. I use HyperFrames to put the chosen clips into a simple sequence, add readable captions, and keep the look consistent. A video with a clear point and a few well-chosen scenes is more useful to me than a pile of flashy effects.

Adapt the idea to the platform. An Instagram Reel needs a quick visual hook. A LinkedIn post needs a personal moment and a takeaway. X works better as one sharp thought or a short thread. Dev.to can explain the workflow and tools. Medium can spend more time on the story and what it means.

Review before publishing. I check the facts, the captions, the tone, and whether the visuals actually support what I’m saying. If an account or publishing integration isn’t connected, the post stays a draft until I publish it through the right channel.
A reusable instruction for the writing step could look like this:

Take my rough idea and identify its one main point. Create a short vertical-video script, then separate drafts for Instagram, LinkedIn, X, Dev.to, and Medium. Keep my first-person voice. Do not invent results, statistics, or experiences. Flag anything that needs checking, and prepare everything for my review before it goes out.

That’s the key distinction for me: repurposing an idea is useful; duplicating the exact same post everywhere usually isn’t.

What’s the “best” way to automate social media?
I don’t think there’s one best tool for everybody. The best setup is the one that removes the repetitive work without flattening your voice.

For my workflow, the simple version is:

*Idea → Claude Code skill → visuals with Higgsfield → assembly with HyperFrames → platform-specific drafts → Instagram through Ootto → review and learn.
*
I’d automate the repeatable parts first: outlining, formatting, captions, resizing, and preparing variations. I’d keep a human checkpoint for the things that represent me: the opinion, factual claims, sensitive replies, and the final decision to publish.

There’s also a practical reason not to assume every platform can be wired up in the same way. Publishing access and permissions differ. For example, LinkedIn’s member-posting API uses OAuth and requires the relevant posting permission. 3 So automation is partly about good prompts—and partly about using integrations that are actually authorized for the account.

Where I think the industry is going
I think we’re moving from “AI, write me a caption” toward small personal content systems that can understand a brief, create a few formats, help publish them, and learn from what people respond to.

That will make video and social content production much more accessible. It will also make generic content easier to produce in huge amounts. When everyone can generate a polished clip, polish alone won’t be the advantage. A clear point of view, real context, taste, and trust will matter more.

The creators I expect to stand out won’t necessarily be the ones who automate the most. They’ll be the ones who use automation to show up more consistently while still sounding like themselves.

That's a conversation, If you guys lives in/near Nagpur, come to DevFest Nagpur 2026 Organized by Google Developer Group Nagpur (GDG Nagpur) on 12th and 13th of Dec 2026, we will have lot more tech conversation there.

AI can help me show up more often. It shouldn’t decide what I stand for.

**_

Note: this post is written and posted by Ai. Ai can make mistakes
_**

New Year, New You Portfolio Challenge - Pranay Gajbhiye

Pranav Shah — Sun, 01 Feb 2026 08:36:08 +0000

This is a submission for the New Year, New You Portfolio Challenge Presented by Google AI

About Me

Hey everyone! My name is Pranay Gajbhiye. I'm 20 years old from Nagpur, India, and I'm currently pursuing a B.Tech in CSE 2nd Year. I like finance and quant trading, and since I'm from a CSE background, I choose to make a mix-and-match of "Design x Trade" and make my portfolio in the theme of my startup (BlackObsidian). We Are Looking For Investors and VCs. If Any Reading This, Please Do Consider BlackObdiain

About Black Obsidian

Yes, I'm also promoting my startup, BlackObsidian, which is an AMC. Earlier it was a quant firm; The BlackObsidian is currently having a CAGR of ~11% and $42K AUM. We are going to move to the UAE for global investors and markets soon this year.

Honestly

I know the Site is bit fancy; since This should be a Portfolio, I somehow used AI 100% to make this site. Also, If any One Of you knows of any startup events, do let me know. In comments

Portfolio

I didn't want to build just another static portfolio; I wanted to create a journey. Black Obsidian invites you to explore interactive 3D scenes, play with real-time data visualizations, and have intelligent conversations with an AI assistant powered by Google's Gemini AI. It's a reflection of my passion for pushing the boundaries of what web experiences can be.

Try these features:

🎮 Innovation: Explore the interactive 3D crystal scene (drag to rotate!)
💬 AI Integration: Chat with the AI assistant about my projects (powered by Gemini Pro)
📊 Real-time Data: View live Bitcoin market data and quantitative order books
🎨 UX: Experience smooth GSAP animations and responsive design

🚀 Live Demo

How I Built It

The core of my submission is the fusion of 3D WebGL graphics with Generative AI. I built this project with a focus on performance, scalability, and modern best practices.

Tech Stack:

Frontend: Next.js 15, React 19, TypeScript
3D Graphics: Three.js, React Three Fiber, @react-three/drei
AI: Google Gemini API (gemini-pro)
Data & Backend: PostgreSQL (Neon), Binance API, D3.js, Recharts
Styling & Animation: Tailwind CSS, GSAP, Framer Motion
Deployment: Google Cloud Run, Docker

Gemini AI Integration

The AI Chat Assistant isn't just a chatbot; it's an integrated guide. By using Google's Gemini API with custom system instructions, I created a persona that understands my specific skills, project history, and professional background.

I used the gemini-pro model for its superior context handling. Here’s the core API route that manages multi-turn conversations:

// app/api/gemini/chat/route.ts
import { GoogleGenerativeAI } from '@google/generative-ai';

const genAI = new GoogleGenerativeAI(process.env.GEMINI_API_KEY || '');

export async function POST(req: Request) {
  const { message, conversationHistory } = await req.json();

  // Initialize the model with specific system instructions
  const model = genAI.getGenerativeModel({
    model: 'gemini-pro',
    systemInstruction: `You are an AI assistant for Pusparaj's portfolio. You are helpful, professional, and knowledgeable about his projects...`
  });

  const chat = model.startChat({
    history: conversationHistory,
    generationConfig: {
      maxOutputTokens: 1000,
      temperature: 0.7,
    },
  });

  const result = await chat.sendMessage(message);
  const response = await result.response;

  return NextResponse.json({
    reply: response.text(),
    timestamp: new Date().toISOString()
  });
}

3D & Data Visualization

The Quantitative Trading Dashboard demonstrates real-time data handling. I connect to the Binance WebSocket API to fetch live price updates and visualize the order book using D3.js. This required careful state management to ensure the UI remains responsive (60fps) while processing high-frequency data updates.

Challenges & Learnings

3D Performance: Loading complex 3D models can kill TTI (Time to Interactive). I solved this by using GLTFLoader with Draco compression and lazy-loading the canvas.
Serverless Cold Starts: Connecting to PostgreSQL from Cloud Run can be slow initially. Connection pooling was essential to fix this.
Gemini Rate Limits: To ensure reliability, I implemented robust error handling on the frontend to gracefully manage API limits.

What I'm Most Proud Of

I'm incredibly proud of the integration between the 3D environment and the AI. Having a floating, interactive obsidian crystal that you can explore while chatting with a smart assistant feels really unique.

I also prioritized a premium ("Black Obsidian") aesthetic with Smooth Scrollytelling using GSAP and keeping Accessibility in mind.

This challenge pushed me to combine three distinct disciplines: 3D Graphics, Full-stack Engineering, and AI. I'm proud of the result—a portfolio that feels alive.

If you enjoyed this demo or found the code helpful, please leave a reaction ❤️ and a comment! I'd love to hear your feedback or answer any questions about the implementation.

Open source under the MIT License. Check out the code on GitHub!

Tags : #devchallenge #googleaichallenge #portfolio #gemini

A Neural Network for Algorithmic Stock Trading.

Pranav Shah — Fri, 09 Jan 2026 15:16:32 +0000

Greetings everyone, this marks my inaugural day and post on this platform.

I am still familiarizing myself with the platform's functionalities, but I am delighted to share one of my ongoing personal projects.

This project is known as the AngleOne Option Trading System (AOOTS). I am open to suggestions for a more suitable name, but for now, it is AOOTS.

This system is a neural network designed for trading in the Indian Stock Market, specifically within SENSEX OPTIONS. I am currently utilizing AngleOne as my Indian stockbroker.

I am posting this from my mobile device and am unsure how to link my GitHub profile. Any guidance on this matter would be greatly appreciated.