LM / Creative R&D
AI × 3D production
Concept car rear illuminated by a continuous red taillight
A production artist’s exploration · 2026

Using AI as
a creative tool.

Two experiments in art direction, creative control, and the space between an idea and a finished image.

Research & experimentation

Lucas Medina / 3D production

Explore the study ↓
01 / The question

How do we keep
the artist in the process?

AI raises real questions about creative integrity, authorship, and the future of our work. This study asks a more practical question: how can we use it to fuel our creativity?

A tool for
creative direction.
Guided by an artist.

01 / INTEGRITY

Keep the intent.

Maintain the artistic integrity of human intervention and art direction.

02 / UNDERSTANDING

Learn the nuances.

Test reference data, prompt structure, and the decisions that shape the result.

03 / CONTROL

Make it directable.

Build a process that supports changes, edits, and more consistent results.

02 / The process

From concept to final frame.

A workflow built around references, previs, and a deliberate feedback loop.

01 ↗

Concept

Establish the idea and visual direction.

02 ↗

Reference sheet

Define the design across multiple views.

03 ↗

3D model

Give the subject physical structure in Maya.

04 ↗

Previs

Set camera movement, cuts, and timing.

05 ↗

Prompt

Translate intent and reference priorities with GPT.

06 ↗

Generate

Bring the prompt and references into Higgsfield.

07 ↗

Review & iterate

Adjust the prompt, references, or animatic.

08 ↗

Finish

Regenerate or upscale; inspect the details.

This is a record of the experiments in the deck. Tool names and model references reflect those tests; results are observations from this project.
Black phone concept with orange side button
Case study 01 / Hardware animation

Designed.
Directed.
Generated.

A minimalist phone concept becomes a test of how much control an artist can keep through the full pipeline.

03 / Visual development

Start with a clear design language.

Teenage Engineering’s restrained product photography informed the concept: minimal forms, precise details, and a polished studio finish.

Inspiration / product forms↗
Inspiration / materials↗
Inspiration / detail↗
Read the original concept prompt
“Generate me a phone concept that looks like it was made by teenage engineering, sleek, minimalist and polished. 3 rear facing cameras minimum. I need several detailed views of the phone, orthographic of top, bottom, front, back, left and right. I also need a 3/4 view as well.”

A useful starting point with photographic qualities and the intended industrial aesthetic.

First generated phone concept↗
04 / Modeling

Where automation meets craft.

I modeled the phone in Maya, then tested whether ChatGPT connected through MCP could build from the reference sheet.

~2 hours / Artist-built
Phone model created manually in Maya↗
13m 55s / GPT + MCP
Automated modeling result from the reference sheet↗
The phone was a promising test for automation. A more complex vehicle exposed different limitations later in the study. Timings are from this experiment, not a general benchmark.
05 / The motion guide

Previs is the cornerstone.

The animatic establishes the camera positions, movement, framing, shot duration, and cuts before generation begins.

For me, this is the most important step toward making the result art-directable. It gives the model a definitive guide, even when perfect repeatability remains out of reach.

Watch the simple animation first. Then compare how faithfully the generated result follows it.
Phone / original Maya animatic▶ Film
06 / Reference hierarchy

Give each reference a job.

I use conversational instructions in GPT to develop a detailed prompt for the next model. I call the back-and-forth “AI-r hockey.”

01 / ANIMATION AUTHORITY

The animatic.

Camera, framing, movement, timing, and shot sequence.

02 / DESIGN AUTHORITY

The character sheet.

Geometry, proportions, components, colors, and physical details.

03 / APPEARANCE ONLY

The photography.

Lighting, reflections, surface quality, and photographic finish.

Read the instruction given to GPT
“Using the attached animatic, character sheet and reference image, make me a prompt for Higgsfield using Seedance 2.0 to generate a product showcase video that mirrors the clean and minimal aesthetic of teenage engineering’s product showcase videos. The animatic should be used as a locked guide for each of the shots”
Explore the full animatic-locked prompt
STRICT ANIMATIC-LOCKED PRODUCT RENDERING TASK DEFINITION Perform a photorealistic appearance transformation of the existing product animation from @Video 1. This is NOT a new video creation task, commercial concept, cinematic interpretation, or storyboard generation task. The provided video is the final approved camera animation. Its composition, framing, motion, shot duration, and visual sequencing are already established. Your ONLY task is to reproduce that existing animation with improved product materials, lighting, reflections, surface fidelity, and photographic realism. REFERENCE HIERARCHY — STRICT PRIORITY @Video 1 — ABSOLUTE ANIMATION AUTHORITY. Controls all camera positions, perspective, focal framing, object movement, scale, composition, timing, and visibility. Reproduce the supplied animation as closely as possible frame by frame. @Image 2 — ABSOLUTE PRODUCT DESIGN AUTHORITY. Defines the precise geometry, proportions, camera configuration, buttons, seams, screws, component placements, markings, colors, and industrial design of the phone. @Image 1 — PHOTOGRAPHIC APPEARANCE REFERENCE ONLY. Defines the desired surface quality, material response, realistic lighting, microtextures, reflection behavior, and image finish. Do not copy its camera angle or composition. When references conflict, preserve the camera animation from Video 1 and the physical product identity from Image 2. ANIMATION AND COMPOSITION LOCK Reproduce exactly the existing shot shown in Video 1. Preserve the original animation's progression from first frame to last frame. Maintain identical camera orientation, perspective, trajectory, and movement speed. Maintain the phone's silhouette, placement, scale, rotation, and cropping throughout the sequence. Preserve every intentional pause, static hold, and subtle movement. Do not add camera movement to otherwise static moments. Do not introduce any additional shots, cutaways, inserts, transitions, or alternate viewing angles. Do not recompose the subject to achieve a more cinematic image. Do not extend the action or invent new content before or after the supplied shot. There must be exactly ONE continuous shot, reproducing the shot contained in the video reference. PRODUCT IDENTITY LOCK The product is a minimalist, precision-manufactured black smartphone with distinctive exposed fasteners, a three-camera rear assembly, modular panel boundaries, a precisely positioned orange side button, and restrained industrial markings. Preserve every physical feature as defined by Image 2. Do not redesign, simplify, reposition, recolor, duplicate, or remove any component. Maintain accurate lens dimensions, button locations, housing thickness, corner radii, chamfers, seams, ports, screws, typography, and branding. Any geometry not clearly visible in Video 1 must be inferred exclusively from the design reference. PHOTOGRAPHIC APPEARANCE TRANSFORMATION Apply an ultra-premium minimalist studio product-photography finish, inspired by the restrained visual sophistication of Teenage Engineering product campaigns. Lighting: Extremely soft, large-area diffused studio lighting with subtle directional illumination, carefully controlled reflections, elegant tonal gradients, delicate contact shadows, and natural ambient occlusion. Materials: Physically accurate fine-textured black anodized aluminum, satin-finished housing surfaces, precisely machined edges, refined camera lens glass, subtle optical reflections, and beautifully controlled surface microtexture. Environment: Neutral light-gray seamless studio background, soft warm-gray ambient illumination, natural tonal separation, clean highlights, controlled black levels, and no distracting environmental elements. Photographic quality: Crisp industrial geometry, convincing surface texture, realistic lens optics, subtle specular response, excellent dynamic range, sophisticated global illumination, and exceptionally clean photographic fidelity. The final image should resemble an immaculate high-end commercial product photograph rather than an obviously computer-generated render. ABSOLUTE EXCLUSIONS NO new shots. NO extra camera angles. NO cinematic inserts. NO unnecessary transitions. NO invented camera movement. NO additional object rotation. NO reframing. NO creative reinterpretation of the animation. NO geometry changes. NO added or missing buttons. NO altered lens arrangement. NO invented details. NO distorted typography or logos. NO unintended material changes. NO dramatic lighting, artificial bloom, aggressive depth of field, lens flares, exaggerated reflections, excessive contrast, stylized effects, or science-fiction visual treatments. FINAL DIRECTIVE Treat the provided video as a locked animation plate. Retain its original shot structure and movement. Change only the product's photographic appearance and environmental lighting. The success criterion is fidelity to the reference animation, not the creation of new cinematic imagery.
The aim: reproduce the existing animation with improved appearance. The prompt asks for no new shots, camera angles, or product redesigns.
07 / Fidelity vs. interpretation

When the model goes rogue.

More references can improve the look and introduce more drift. In this test, the same prompt with a higher resolution and extra appearance references produced a different interpretation.

First result / 480p source▶ Film
Additional references / 720p source▶ Film
See the additional appearance references
Additional lighting and material reference↗
Additional finish and detail reference↗
A visually stronger result can still be less faithful to the animatic. Evaluate appearance and shot accuracy separately.
08 / Finishing

More detail.
But is it the right detail?

Upscaling can invent texture or sharpen away the depth of field. The test is whether it preserves the photographic qualities of the original.

Original generation / 720p source▶ Film

Inspect the interpretation.

Look for invented surface detail, over-sharpening, and changes to focus. Prompted upscaling can offer another way to steer the finish.

Start with more information.

My takeaway: generating at a higher resolution before upscaling gives the model less missing information to invent.

Web videos are playback copies capped at 1080p. Labels describe the source files in the Keynote. Use the original 4K files for pixel-level quality judgments. Topaz was listed as another option in the deck, but no Topaz result was included.

Silver concept car in a dark studio
Case study 02 / Automotive showcase

Same process.
More complexity.

A concept car becomes the next test: apply the phone workflow, automate more of the process, and find the limits.

09 / Concept development

One sketch.
A consistent visual identity.

I started with my own side-view sketch, asked GPT for additional views, then worked toward a simpler, more consistent reference sheet.

Original concept sketch / side view↗
Read the simplification prompt
Can we simplify the body lines and sharp styling of this car? I want the overall silhouette and shape of the car to stay the same. Please create orthographic views, top, front, side, back.
Explore the automotive inspiration
Studio lighting reference↗
Rear design reference↗
Side design reference↗
Silhouette reference↗
Front three-quarter reference↗
Dark studio photography reference↗
10 / Automation test

Complexity exposes the gaps.

Could GPT model the vehicle through Maya MCP? This was a much harder test than the phone. The automated attempts made the limits visible.

Automated vehicle modeling / iteration contact sheet↗
Final car modeling views from the experiment↗

“AI won’t be replacing me … for now.” The study’s takeaway from the modeling phase: a consistent reference sheet helps, but it does not remove the need for an artist’s technical judgment.

11 / Establish the shots

Back to the animatic.

The same foundation: a clear motion guide for the generated film.

The vehicle’s previs establishes the shots before materials, lighting, and photographic finish enter the process.

Automotive / original previs animatic▶ Film
12 / Art direction

Separate the look
from the design.

One image controls the photography. Another defines the vehicle.

The deck’s automotive prompt spells out the studio environment, overhead softbox, negative space, metallic response, and strict preservation of the concept car’s proportions.

This prompt is the still-image art direction shown in the deck; it establishes the photographic intent for the automotive work.
Photography and environment reference↗
Explore the full automotive photography prompt
Create an ultra-photorealistic automotive editorial photograph using the two provided images. IMAGE REFERENCES Image 1 — Photography and environment reference: Use this image as the primary reference for composition, lighting, exposure, atmosphere, color grading, and overall photographic aesthetic. Image 2 — Vehicle design reference: Use this image as the exclusive reference for the car's design, shape, geometry, proportions, materials, and details. COMPOSITION & ENVIRONMENT Recreate the exact studio environment and visual atmosphere of Image 1, replacing the Porsche with the silver concept car from Image 2. The scene takes place inside an enormous, nearly pitch-black industrial photography studio with a massive rectangular overhead softbox suspended above the vehicle. The softbox is recessed within a dark rectangular housing, creating a dramatic architectural lighting feature in the upper portion of the composition. Position the concept car near the bottom of the frame, maintaining the same relative scale, camera distance, negative space, and visual weight as the original Porsche photograph. Photograph the vehicle from a low, slightly elevated rear three-quarter perspective, similar to Image 1, revealing the side profile and rear of the vehicle. Maintain the same camera height, approximate focal length, framing, and perspective. LIGHTING Match the lighting in Image 1 as closely as possible. The massive overhead softbox is the dominant illumination source, producing soft, directional, naturally diffused white light that falls gently across the car. The vehicle should remain predominantly in shadow, with subtle illumination revealing its silhouette, sculpted surfaces, window contours, wheels, and rear bodywork. Create delicate, physically realistic specular reflections across the silver metallic paint. Avoid strong artificial rim lights, exaggerated reflections, or excessively bright surfaces. The studio floor receives a narrow pool of soft illumination beneath the vehicle, while the surrounding environment fades into almost complete darkness. Allow atmospheric haze to subtly reveal the light falling through the space, with realistic light scattering and very gentle volumetric illumination. PHOTOGRAPHIC REALISM The result must look like a real photograph of a physically manufactured concept car, not a CGI render or digital illustration. Emulate premium automotive editorial photography: Natural optical imperfections and subtle film grain. Deep, nuanced blacks with gently lifted shadow detail. Slightly cool, desaturated cinematic color grading. Realistic metallic paint response and imperfect micro-surface reflections. Natural light falloff and physically plausible indirect illumination. Subtle atmospheric depth and realistic contact shadows. Restrained contrast with no artificial HDR appearance. Preserve the moody, mysterious, minimalist, almost architectural quality of Image 1. The image should feel quiet, elegant, cinematic, and extremely sophisticated. VEHICLE ACCURACY — CRITICAL Faithfully reproduce the silver concept car from Image 2. Preserve its exact silhouette, wheelbase, roofline, body proportions, five-spoke wheels, front and rear overhangs, glass canopy, headlight and taillight designs, and all distinctive exterior design details. Use the orthographic views in Image 2 to reconstruct a consistent, believable three-dimensional vehicle from the requested camera angle. Do not borrow any design details from the Porsche in Image 1. Do not redesign, reinterpret, simplify, or embellish the concept car. Do not add new vents, aggressive aerodynamic elements, decorative body lines, or other invented features. FINAL ART DIRECTION The finished photograph should feel almost indistinguishable from Image 1 in terms of mood, lighting, environmental composition, and photographic character. The only meaningful change should be the replacement of the original Porsche with the silver concept car. Prioritize the photographic realism and atmospheric restraint of Image 1 over making the vehicle appear overly polished, brightly lit, or visually exaggerated. Output: Vertical composition matching the original photograph, approximately 3:4 aspect ratio, high resolution, with substantial negative space above the vehicle. Avoid: CGI appearance, showroom lighting, glossy advertising renders, excessive contrast, unnaturally sharp edges, overexposed metallic surfaces, dramatic colored lighting, artificial lens flares, altered car proportions, and any extra objects or distractions.
13 / Generation tests

The same inputs
can produce a different film.

The next test kept the prompt and references the same and changed the resolution. The result still changed. Clearer references, wider shots, and shorter animatic segments may help reduce unexpected invention.

Automotive / first generation▶ Film
The side-by-side comparison is a single synchronized video from the deck, so both versions stay aligned during playback.
14 / The feedback loop

Play a little AI-r hockey.

Generate. Review. Ask for a revised prompt. Generate again.

I feed the result back into GPT, describe what should change, and bring the revised prompt into Higgsfield. Each round is another art-direction decision.

Multiple generations / synchronized comparison grid▶ Film
15 / Selected results

Where the process landed.

Two favorites from the automotive experiment, upscaled to 4K in the original study.

Selected automotive film / 01▶ Film
Selected automotive film / 02▶ Film

Playback copies are optimized to a maximum of 1080p for the website.

16 / What I learned

The potential is real.
So is the work.

AI exceeded my expectations for quality and creative potential. Getting useful results still takes direction, iteration, and a trained eye.

The biggest constraints in these experiments were time, budget, and unpredictability. Generations take time, revisions consume credits, and moving between tools adds work before the final image is ready.

KEEP EXPLORING / 01

Texture refinement.

Improve surfaces while preserving materials, focus, and photographic realism.

KEEP EXPLORING / 02

Masks & mattes.

Find usable ways to support compositing and post-production control.

KEEP EXPLORING / 03

Pipeline automation.

Reduce repetitive handoffs while keeping creative decisions with the artist.

17 / To be continued

The artist is still essential.

“I don’t believe that true world-class design work will ever be a one-click button.”

Lucas Medina

Final concept car / studio photograph↗
Explore the study
Introduction 01—02Hardware animation 03—07Finishing & upscaling 08Automotive showcase 09—15Takeaways 16—17

Explore by scrolling, or choose Present to move one scene at a time. In presentation mode, use ← / → or Space. Press Esc to return to scrolling.

From the original Keynote