DEV Community: Sirajuddin Shaik

Video editor with a built in agent harness

Sirajuddin Shaik — Wed, 12 Aug 2026 14:21:01 +0000

Introducing Snapflow
A video editor with a built in agent harness

Perform Multi agent edits in parallel with Agents , watch and steer agent edits via editor panel

Agents access same set of tool available to you as an editor ,work in same ground visible in editor
Connect the snapflow mcp to your agent , delete repetitive and prototyping tasks to agents , to polish your workflows around editing
Launch an instance via linux , or ask the agent to launch an instance with display

Snapflow mcp comes with tools to launch headless instances , make multi prototyping earlier by isolating projects

Available Starting with Linux
Know more at : snapflow.zealbase.io

Starting Out Content Editing with AI: A Guide to Prompting

Sirajuddin Shaik — Sat, 25 Jul 2026 13:59:31 +0000

You wanted to create a video album consisting of photos from a recent event, combined with nice music to share with your friends, or say to address one of your projects where you're trying to make a presentation, and wanted to take a glimpse of videos from different sources, put them together, and create a nice explanation of your research on top of it. You want to put some facts together from the internet and reference articles or different images, and present these in the form of a video.

Similar editing work can be done using your existing AI models or AI provider subscriptions.

The first idea that may strike is to ask an agent to do that—it's more efficient certainly:
hey, Claude. Do detailed research of this recent movie's explanation of original characters and prepare a report. Make no mistakes.
This prompt is not concise and you may end up with output diverging from your expectations.

Taking the same movie explainer content work, when choosing a voiceover explaining the plot and story context one would like to convey as a narrator,
You can provide some narration to the agent,
explain the whole video scenario, what you wanted to explain about each character when you prepare an audio transcript of that, and ask the agent:

Get me the information of all these characters across the internet and find out facts which are less known. Based on published movie sources, descriptions, or user reviews, filter out facts which are not prevalent in the movie. Let's put together the timestamps where I speak about each character and add visuals specifically close to the original look of characters by checking sources like Wikipedia, etc. Put together a nice sequence matching the audio transcript of the narration.

You can use this form of prompt, which is elaborative and mentions which sources to follow.

This allows the model to make decisions precisely aligned with your expected output. This creates more enriched content. Someone can further elaborate it to specify time ranges, visual colors or animations for text, and specific sizes or properties of visual elements.

When you want images or these details presented as a video for a one-off prompt or one-time content creation, it comes in handy.

Using the provider directly takes a bit of time for research, and you can also ask it to use Python or a specific library to create this video content in such a way.

Here is an example prompt which I just gave to Claude to make use of different geometric measurements and elements, prepare a nice presentation of each, and use math animations to animate these elements over time.

It utilized some Python tools and their timing utilities and animations.
Say, taking out the audio narration part, I did it in a different way instead of providing an audio transcript of the movie.
I gave the whole task to the application to search for all the details itself and prepare an audio transcript, taking it a bit further.

Here is how is the prepared output and it took around an hour with around 5 iterations

The video editing process takes a few minutes to hours, depending upon the length of the content creation task and the depth of the task you have assigned to the agent.

It consumes a chunk of subscription usage limits to some extent,
so it might be good for one-off tasks and a few trials.
Say you do video editing on a regular basis and want to have more control over detailed pictorial aspects of your content. You may want to use specific tools that allow you to see real-time effects of how the video is being processed or edited by the agent. In such scenarios, you need token-efficient methods for the agent to process and edit video.

So there are some ways to do this in an efficient manner, which I am going to explain in my upcoming articles.

Illustrations with Agents - Prompts , Code

Sirajuddin Shaik — Tue, 16 Jun 2026 15:41:38 +0000

Logos and illustrations are pieces of art that one can frequently find themselves looking for, for reasons such as projects, social media posters, etc.

In this case, I needed illustrations to enhance the outlook of a GitHub project.

Naturally, the alignment was to use tools like Canva, Whimsical, or some other tool that can blend images, nodes, and elements to wire them together. To go a step further, one can try using tools that give precise control over edges, geometrical elements, and other items that craft a well-designed logo or illustration.

In an era where AI works well at programming, an idea that struck me was to write programs that draw sketches and generate illustrations, with post-processing by adding colors, etc.

The idea was to give natural geometrical shapes through natural language input to the model, which then goes on to craft the art.

This also allows trying different versions of art through parallel agents, which can enhance the art through a few more ideation and iteration processes.

To try this out, I started with a logo. After a few runs, and sometimes using parallel prototyping of the design, I arrived at this.

Logo For Graft

The illustration was also created by following a similar pattern, although this used HTML and SVG to capture and create short animation-style content.

The architecture used to prototype this was:

It allows one to quickly try different variations of art. This was a brute-force way of generating images, and enhancement of the stack can lead to better and more desired output.

As I had a side project for GitHub, I was trying to create logos and illustrations to present convey information via Illustration on GitHub, This started as a side project itself exploring some possiblities of art through words , aks prompts