Founding Applied Computer Vision Engineer (Identity Preservation, Pose Transfer & Character Consistency)

Build something billions of people will use:

Billions of people take photographs. Almost everyone has experienced the same frustration: the moment happened, but the photograph did not capture them at their best.

We are building Spotted—a UK-based AI imaging company creating technology that turns ordinary phone photographs into extraordinary, professionally captured images.

This is not a learning role

We are looking for an exceptional senior applied computer-vision or generative-imaging engineer who has already worked on closely related problems—or who can demonstrate that they can solve this one quickly.

Please apply only if you can independently:

- understand the problem from first principles;
- evaluate existing methods honestly;
- build and combine the right approaches;
- produce a working demonstration quickly;
- measure identity and geometry preservation properly;
- show failures rather than hiding them;
- turn an experimental workflow into defensible technology.

You will not receive a pre-written technical recipe. You will help determine what should be built. The problem you will own

Our current workflow uses a generative image model to create alternate poses of a real person in the same scene. The new composition and body pose may be excellent, but the generated face often becomes a slightly different person. It may look similar to the original subject while changing their underlying facial structure. Your job is to build a system that:

1. Retains the generated target pose, expression, gaze, clothing and scene.
2. Restores the real person’s identity and facial geometry.
3. Preserves the jaw, cheeks, eyes, nose, chin, facial outline, natural asymmetry and age cues.
4. Integrates the reconstructed face or head naturally with the target lighting and image quality.
5. Does not silently beautify, reshape or standardise the person.
6. Leaves the body, clothing, hands and background untouched.
7.



Detects uncertain cases and rejects them instead of returning the wrong person.

“People can recognise them” is not our definition of success. The output must still be the actual person.

What you may explore

We are open to the strongest technical approach. Relevant directions may include:

- identity-conditioned diffusion and reference adapters;
- character-consistency systems;
- person-specific or identity LoRAs trained from consented galleries;
- face-ID conditioning;
- conventional face-swapping baselines;
- face or head-level reconstruction;
- dense facial landmarks and pose-normalised geometry;
- 3D-aware facial reconstruction;
- ControlNet and structural conditioning;
- segmentation, matting and controlled compositing;
- geometry-aware identity evaluation;
- multi-reference identity enrolment.

We have access to multiple consented photographs of each user. You should think beyond single-image face swapping and consider how a person’s gallery can become a strong identity reference. A LoRA, face swap or high similarity score is not automatically a solution. The method must preserve actual identity geometry across new poses.

What we need from you

You should bring solid, demonstrable experience in several of the following:

- applied computer vision and generative imaging;
- PyTorch and production-quality model experimentation;
- diffusion or image-to-image systems;
- character consistency across pose and expression changes;
- face swapping or identity-conditioned generation;
- person-specific LoRA training;
- face restoration or head reconstruction;
- dense landmarks, 3D face models or geometry-aware guidance;
- ControlNet, depth, pose or structural conditioning;




- segmentation, matting and seamless compositing;
- image-quality and identity evaluation;
- reproducible experimentation and failure analysis.

Direct evidence matters more than credentials. A PhD, paper or prestigious employer is valuable only if you can explain what you personally built and demonstrate that you can deliver this system.

What exceptional performance looks like

Within the initial proof period, you will:

- reproduce and evaluate relevant existing baselines;
- compare face swapping, identity conditioning, person-specific LoRA and geometry-aware methods;
- build a working identity-preserving pose-transfer demonstration;
- test it on a locked set of frontal, three-quarter and profile poses;
- report every result—not only selected successes;
- measure identity, facial geometry, target-pose retention and untouched-region preservation separately;
- document runtime, compute requirements and cost per image;
- identify hard failures and create a rejection or fallback strategy;
- recommend clearly what we should build, buy, combine or abandon.

We value speed—but not speed created by hiding failures.

Who you are

You are likely someone who:

- wants to solve a technically difficult problem with enormous consumer potential;
- is energised by ownership rather than waiting for detailed instructions;
- moves rapidly from research to working demonstrations;
- is intellectually honest about what does and does not work;
- cares about the difference between an impressive demo and a repeatable product;
- wants to help build the company, not merely complete assigned tickets;
- is comfortable working directly with founders;
- wants meaningful equity and influence over the technology you create.

Apply because you have already developed the necessary depth—or because your prior work proves that you can solve this quickly. If that might be you, we want to hear from you.

Contact us at- Email: [email protected], Whatsapp: +919827123412

📌 Applied Computer Vision Engineer (India)
🏢 Spotted App
📍 India