DEV Community: Jesus Lopez

Designing AI Camera Workflows Around a Single Reference Image

Jesus Lopez — Mon, 05 Oct 2026 08:18:05 +0000

A reference image gives a creative project a starting point, but it rarely answers every visual question. A product designer may need to see the same object from above. A character artist may need a side view. A filmmaker may want to test what happens when the camera pushes toward a still scene.

These sound like variations of one task, but they produce different kinds of output. A new viewpoint is a still image. A camera path is a video. A turntable is a sequence of views. A scene template may combine multiple references into one performance. Keeping those workflows distinct makes it easier to choose the right tool and evaluate the result.

I organized 3D Camera Control around those differences. Here is how I think about the main workflows, what each can help with, and where AI-generated results need a closer look.

1. Exploring a new viewpoint

For a still image, the first decision is usually where the viewer should be. Rotation changes which side of a subject is visible. Elevation moves the viewpoint above or below it. Distance changes the framing.

A useful experiment changes one of these at a time. If a front view is your starting point, try a side angle at roughly the same distance. Then return to the original rotation and raise the camera. This gives you a clearer sense of which decision made a composition more useful.

Different subjects call for different checks:

A face angle change can help explore a three-quarter or profile portrait. Compare facial features with the source image rather than judging the new composition alone.
An AI pose change explores body position. The question is whether the new stance works while the person, outfit, and setting remain recognizable.
A zoomed-in image can focus attention on a detail, while a zoomed-out image explores a wider frame and the surrounding scene.

When one view is not enough, the output can become a small set. A character turnaround provides front, side, or back references for discussing a design. Multiple product views help compare angles for a concept board or a planned photo shoot. A 3D image spinner lets viewers inspect a sequence of 2D frames around a subject.

The word “3D” can be misleading here. These generated views are images, not a measured 3D model. If the source does not show the back of an object, AI has to infer it. That makes the results valuable for exploration, but important geometry, labels, and identity details still need human review.

2. Moving the camera through time

A video adds another decision: how does the viewpoint travel between the opening and closing frames? A push-in, an orbit, and a fast pan can all start from the same image but create very different effects.

The camera control for video workflow focuses on planning a camera path with keyframes. The AI camera movement workflow provides movement references such as dolly, pan, orbit, and aerial motion that can be refined before generation.

Some movements are easier to evaluate as focused examples:

A dolly-in shot moves attention toward the subject.
A whip pan explores a rapid change in direction.
An AI drone shot explores an elevated or aerial camera path.
A face zoom or eye zoom tests a more specific framing effect.

For moving shots, I look beyond whether an individual frame is attractive. Does the subject remain recognizable throughout the clip? Does the motion communicate the intended change in attention? Does the background behave plausibly as the camera moves? Watching the full result matters more than checking a thumbnail.

3. Starting from a character or scene concept

Sometimes the creative question begins before the camera is chosen. An AI influencer generator can help explore a character concept from a description or photo and build visual references for later work.

A template solves a different problem: it fixes much of the scene so that the user can focus on the inputs. For example, Hotel Lobby AI takes two separate performer photos and places them in an orange recording-booth duet. The template defines the setting and performance direction; the generated video still needs to be reviewed for facial details, gestures, and timing.

Templates can make an experiment quicker, but they do not remove uncertainty. Clear source images and a review of the finished output remain important, especially when a result includes identifiable people or details that should stay consistent.

A useful review process

Across all these workflows, I find it helpful to decide what success means before generating. Write down the subject details that must survive, the viewpoint or movement you want to test, and how the result will be used. Then compare the output with the reference.

A product concept image may be enough for a mood board but unsuitable for describing the exact item being sold. A character turnaround may expose design questions without being consistent enough for final modeling. A video may communicate a camera idea even if individual frames need another pass.

That distinction is the point of these tools: they help explore and communicate visual options. The generated result is a reference to inspect and refine, while the final decision about accuracy and usefulness stays with the creator.

I Built GPT Image 2.5: AI Image Generation with Both Classic Prompts and a Chat Agent

Jesus Lopez — Thu, 10 Sep 2026 15:52:51 +0000

I recently built GPT Image 2.5, a web-based AI image generation and editing workspace:

https://gpt-image-2-5.org/

Most AI image tools follow the same pattern: choose a model, write a prompt, select a size, click generate, and wait for the result. That workflow is still useful, so GPT Image 2.5 supports the classic image generation flow.

But I also wanted to explore a more natural way to create images: generating through conversation.

Two ways to create images

GPT Image 2.5 currently supports both workflows:

Traditional prompt-based image generation
Conversational image generation with an AI agent

The traditional mode is good when you already know what you want. You can write a prompt, upload a reference image, choose the model and output size, then generate or edit the image directly.

The AI agent mode is different. Instead of filling out everything manually, you can chat with the agent about what you want to create. For example, you can describe a product photo, a poster idea, a portrait style, or a marketing visual, and the agent helps turn that conversation into an image generation task.

This makes the workflow feel closer to working with a creative assistant:

You can start with a rough idea
Ask the agent to refine the direction
Adjust the style through conversation
Explore multiple visual concepts
Generate images without needing to write a perfect prompt from the beginning

Supported models and use cases

The workspace includes Flare, Sunburst, and GPT Image 2.

Flare is useful for fast creative exploration. Sunburst is better suited for more detail-focused editing. GPT Image 2 remains available for workflows built around the previous generation.

Some example use cases:

Product visuals
Portraits
Posters
Campaign concepts
Social media images
Reference-based image editing
Creative visual experiments

Why add an AI agent?

A lot of people do not naturally think in “prompt format.” They may know the result they want, but not how to describe it in a way that works well for an image model.

The agent helps bridge that gap.

Instead of forcing users to write a polished prompt immediately, the site lets them describe the idea conversationally. The agent can help shape the request, clarify the visual direction, and guide the generation process.

For developers and makers, I think this is an interesting direction: AI image tools do not have to be just prompt boxes. They can become interactive creative workspaces.

What I am improving next

The project is still early, and I am working on improving:

The chat-to-image workflow
Better prompt refinement
More image editing options
More examples and templates
A smoother model selection experience

I would love feedback from developers, designers, marketers, and creators who use AI images in real workflows.

Website: https://gpt-image-2-5.org/

I Built an AI Tool That Turns Flat Images into Editable Layers

Jesus Lopez — Mon, 31 Aug 2026 10:05:09 +0000

I Built an AI Tool That Turns Flat Images into Editable Layers

A flat image can look complete, but it is often difficult to edit.

If you want to move a product, remove a shadow, replace a background, edit text, or reuse part of a composition, you usually have to recreate the image structure manually in Photoshop.

That was the problem we wanted to solve with Image2Layer.

What Image2Layer does

Image2Layer uses AI to analyze a flat JPG, PNG, or WebP image and separate it into meaningful editable layers.

Depending on the image, the result may include:

Main subjects
Foreground objects
Product elements
Text and logos
Shadows
Decorative elements
Background areas
Reconstructed regions behind removed objects

Each layer keeps its transparency, so it can be moved, hidden, edited, or recomposed independently.

Export formats

After processing an image, you can export:

Transparent PNG layers
A ZIP package containing the layers
A layered PSD file for Photoshop and other creative workflows

The goal is not to recreate the original source file perfectly. Instead, Image2Layer provides a useful editable starting point from an image that would otherwise be difficult to work with.

Who is it useful for?

Some possible workflows include:

Retouching product photos
Creating new e-commerce compositions
Reusing poster and marketing assets
Preparing images for animation
Separating characters and objects from illustrations
Rebuilding backgrounds
Moving from an AI-generated image to an editable design

How it works

Upload a JPG, PNG, or WebP image.
Choose the layer detail level.
Add separation notes if certain elements are important.
Generate the layers.
Review and download the result as PNG, ZIP, or PSD.

You can try it here:

Try Image2Layer

What we are still improving

AI separation is not equally easy for every image. Fine details, transparent objects, reflections, complex typography, and heavily overlapping elements can still be challenging.

We are continuing to improve:

Layer boundaries
Background reconstruction
Text and logo separation
PSD export quality
Control over the desired layer structure

If you work with image editing, design, or creative automation, I would be interested to hear what kind of images you would want to split into layers.