Join Dashverse Research

Build the Future of Generative Content Creation

Welcome to Dashverse — where we don't just use generative AI, we push it to its limits.

We are a small, fast-moving team building powerful tools for the next generation of storytellers. Our products — like Frameo.ai and Dashtoon Studio — enable creators to instantly turn ideas into high-quality videos, comics, and more using AI. If you're someone who thrives on solving deep technical problems and wants to see your work used by millions, this is your playground.

The Role: Research Engineer

This isn't your average research role.

At Dashverse, your work directly impacts creators. You’ll get to see the delight on a user's face when a tool you've built helps them bring their imagination to life — in seconds, not hours. This is a rare opportunity to move fast and shape the future of generative media.

You're not here to write papers and wait six months for reviews. You're here to:

• Prototype bleeding-edge generative models.
• Ship tools that change how stories are made.
• Build the infra and internal tools to move 10x faster.

What We're Solving

We’re building real tools for real creators — and that means diving deep into unsolved problems at the frontier of generative media.

Advancing Visual Quality with Diffusion

Diffusion models are powerful, but raw outputs often fall short. Whether it's a single comic panel or 120 frames of an AI-generated video, visual quality matters. We're focused on pushing fidelity, coherence, and artistic control to new heights.

Prompt: A 1950s All India Radio broadcaster reading news in Hindi at a wooden microphone, headphones on and script in hand under soft studio lights.

Flux Dev	Dashtoon Model

Controllability at the Core

We're developing creative tools, not black boxes. That means giving creators meaningful control over outputs:

• Text + image multi-modal prompts.
• Conditioning on pose, expression, scene layout, and panel structure.
• Temporal control for frame-to-frame consistency in videos.
• Spatial control for region-specific edits.

Character Identity & Reusability

Characters aren't just faces — they're consistent, recognizable identities. We're building:

• ID-consistent inpainting to fix or change parts of a character without losing fidelity.
• Low-data personalization methods (e.g., style or identity adapters).
• Seamless multi-angle rendering of the same character across scenes or panels.

This enables creators to design once and reuse intelligently — with style and consistency.

Scene-Aware Object & Clothing Transfer

We're building pipelines to enable asset-level swaps (outfits, props, etc.) while preserving visual integrity and scene coherence.

Key techniques include:

• Semantic segmentation with feature-level transfer.
• Training-free approaches using adapter stacking.
• Support for both photoreal and stylized (anime, comic) domains.

Base Image	Clothing Image	Final Image

Training-Free Customization & Distillation

We're exploring fast, lightweight personalization methods that don’t require full retraining:

• Tuning-free ID adapters and style injectors.
• LoRA + ControlNet workflows with zero additional training.
• Fast model distillation for mobile inference and real-time generation.

These innovations unlock high-speed, high-fidelity use in production creator tools.

If any of this sparks ideas, feel free to explore one of our sample problems — or fork the repo and show us what you’d build differently.

What You’ll Work On

Here are a few recent open-source drops from our team:

• Keyframe LoRA: Video generation using only keyframes
• Tuning-free ID-consistent inpainting
• DashTailor: Object transfer for AI-generated comics
• Adversarial Diffusion Distillation
• DashAnimeXL: Stylized diffusion for comic-style animations

Who We’re Looking For

We don’t care about your resume or degree. We care about what you’ve built, trained, or broken and fixed.

You might be a good fit if you:

• Have trained or fine-tuned models like Stable Diffusion, AnimateDiff, or StyleGAN.
• Are comfortable with PyTorch, LoRA, ControlNet, DreamBooth, or IP-Adapters.
• Can manage multi-GPU training jobs, optimize inference, and debug issues fast.
• Prefer to move quickly and own the entire stack from training to UI.

Bonus if you:

• Have written clean infra or tooling that helped teams scale.
• Have contributed to open-source or built an impressive side project.
• Have a strong perspective on how generative models should evolve.

Perks

• Dedicated compute access with high-end GPUs.
• Fixed-band compensation based on submission quality.
• End-to-end ownership: from research to production to user experience.

How to Apply

• Clone this repo.
• Attempt one of the sample problems — or fork it and show us something better.
• Email your submission or proposal to: soumyadeep [at] dashtoon.com

No Leetcode. Just real problems and real builders.

If you're the kind of person who wants to own the full stack — from fine-tuning to infra to user-facing tools — come build with us.

Dashverse — Dream it. Generate it. Share it.