Esha Sharma

AI Workflow Researcher. Turning broken nodes into working. Current Focus: Wan 2.2 Optimization. Get the files:

[Esha Sharma @aistudynowcom]

I just tested the new Kera 2 Outfit Swap LoRA, and the results are amazing.
Yesterday, I made a video to show you how this model works. After more testing, I believe this is the best outfit transfer LoRA available today.

[Esha Sharma @aistudynowcom]

@invideoOfficial https://t.co/ceXwjA3ij5

[Esha Sharma @aistudynowcom]

Did you see my new YouTube tutorial on making a consistent AI short film?
It is usually very hard to keep One environments connected in long videos.
I finally figured out how to achieve that!
I built this entire workflow using @invideoOfficial.
Everything was done through a simple chat.
I did not manually write the story, prompts, storyboard, or video plan.
I saved my exact setup into a file for you. Here is what is inside:
The complete video plan and prompts
The consistent character agent
The @invideoOfficial workflow settings

[Esha Sharma @aistudynowcom]

Do you like scary stories? I just finished making a 1-minute AI horror video.
The best part is how consistent the environment looks.
The background stays perfectly stable throughout the entire story.

[Esha Sharma @aistudynowcom]

I just finished animating a fun donut heist and police chase.
The best part is how consistent the environment looks.
There are zero overlapping issues or duplicate items in the bakery.

[Esha Sharma @aistudynowcom]

Made an anime-style summoning scene.
She performs a beautiful water dance to summon a giant water dragon.
The best part is the character consistency.
When the dragon strikes with full power, there are zero overlap issues. Not a single frame was generated wrongly.

[Esha Sharma @aistudynowcom]

The hardest model I’ve worked with so far is HiDream-O1-Image.
Yes, I can cherry-pick good results and make a video.
But my real goal was different:
I wanted to understand what actually makes this model work. Fix the Plastic Skin

[Esha Sharma @aistudynowcom]

Remember that fighting scene I showed you yesterday? The story continues, but it gets dark.
Our hero trained hard and fought back again. Sadly, the villain is still too powerful and wins. 😉

[Esha Sharma @aistudynowcom]

@waller67063 using seedance 2.0, but can also make this using ltx 2.3

[Esha Sharma @aistudynowcom]

I just created a Fighting Scene! The result is stunning, consistent, and the physics are also correct.

[Esha Sharma @aistudynowcom]

@nlanphu I generated this using seedance 2.0, but can also make this using ltx 2.3

[Esha Sharma @aistudynowcom]

I generated this exact video result in LTX 2.3 with just 8GB of VRAM.
I have a 32GB system, so I reserved 24GB to test how it performs on an 8GB setup. It renders very fast!
The full workflow is coming soon.
What to expect in the workflow: • Fast render speeds on 8GB VRAM • Simple and easy setup • Built mostly with native nodes
Watch the result below and follow me so you do not miss the workflow when it drops!

[Esha Sharma @aistudynowcom]

I just Generated this video using the Seedance 2.0 inside @YouArtStudio . I am honestly thrilled with the final result.
You can Now Generated this Same Result With LTX 2.3 Too and it is Possible now

[Esha Sharma @aistudynowcom]

I just tested the Seedance to build a luxury "Get Ready" fashion storyboard. I wanted to achieve something incredibly difficult. We needed to keep one female subject perfectly consistent while she builds a complete outfit across 16 different fashion moments.
Here is the exact step-by-step progression we used:
Base: White tee and trousers
Layers: Shirt and blazer
Accessories: Watch, necklace, and belt
Final Details: Sunglasses, shoes, and a bag
The Agent successfully maintained her exact visual identity through every single layer.
Created with the Seedance 2.0. Credit: @YouArtStudio

[Esha Sharma @aistudynowcom]

I noticed Z-Image Omni showing up inside ComfyUI today. The comfyui includes the new node, but we have a slight problem. The developers still list the official Z-Image-Omni-Base model as "To be released."
I definitely wouldn't call this a full launch yet. ComfyUI updated their platform before the actual Omni checkpoint became available for you to download.
This specific model changes your entire workflow. We expect it to do much more than standard text-to-image generation.
Here is exactly what the new system handles:
Creates new images directly from your text prompts.
Processes advanced image editing inside the exact same node.
Eliminates your need to switch between different generation and editing models.