In Part 1, I walked through three sessions and 46 images building a temporal portal concept that scored 92.60 and became my Adobe "How I Firefly" competition entry. A contemporary loft, a Victorian library, a gravity well pulling objects in a mathematical spiral through a luminous circular portal. The formula worked.

The obvious question after that: does it scale?

One portal showing one time period is a composition problem. Two portals showing two different time periods -- past and future simultaneously, with the present as the fulcrum between them -- is a different kind of problem entirely. Two gravity sources. Two lighting temperatures competing for the same frame. A concept that could collapse into visual noise or resolve into something more resonant than the original.

One session. 30 images. The answer turned out to be the second thing.

Before writing a single test prompt, there is a structural question to answer: where does the viewer stand?

In the original temporal portal formula, the answer is clear. You are in the present. The portal shows you the past. The gravity distortion is pulling things toward the past. There is one direction of tension, one temporal destination, one story.

Three eras break that geometry. If there is a portal to the past on the left and a portal to the future on the right, the present room is no longer an anchor. It is a crossroads. The viewer is being pulled in two directions simultaneously. The physics have to account for two gravity sources, the lighting has to show two competing temperatures, and the concept has to hold together as a readable image without a single dominant focal point.

That is what I was testing. Not just whether Firefly could render three time periods -- I assumed it could -- but whether the concept would produce something more visually complex and interesting than the two-era original, or just more cluttered.

The first variation was baseline: contemporary loft, Victorian library on the left, a near-future minimalist space with bioluminescent panels and clean architecture on the right. No spiral, no figure. Just the core concept to establish whether three eras could coexist in a single frame.

Three of the four images produced a white void where the future portal should have been.

This is a failure mode I have documented before: the specificity problem. It is more acute for future eras than historical ones. When Firefly renders a Victorian library, it has an enormous reference base. Bookshelves. Gas lamps. Leather chairs. Mahogany desk. Every one of those words resolves to a clear visual. When I wrote "bioluminescent panels and clean architecture," I gave it concepts rather than things. The model filled the gap with brightness. Absence rather than presence.

The fix required a complete rethink of what a convincing future actually looks like in photographic terms. Not chrome. Not floating geometry. Not cold blue light and clean lines -- that is the 1980s version of the future, rendered a thousand times, immediately generic. I went the other direction: organic, warm, specific.

The future portal language that solved it:

One piece of technology among natural materials. The rest is craft and living things.

That rewrite solved the problem across every remaining variation. The moss walls showed up in every image. The warm timber surfaces appeared consistently. The single screen signaled future without triggering generic sci-fi rendering.

Going warmer on the future also broke the obvious warm/cold shorthand that Firefly defaults to. Which meant the model had to distinguish past from future through architectural specificity rather than just color temperature. The images were better for it.

The lesson that transfers: the future is just as hard to describe as the past unless you ground it in materials. "Clean and futuristic" is not a description. Curved timber, living walls, and one screen is.

The mathematical spiral was the single most important technical element in the original formula. Objects following a precise curved trajectory demonstrating gravitational physics -- not random floating, not decorative drift. The spiral is what makes the image look like a physics demonstration rather than a magic trick.

Testing whether that technique survived the introduction of a second gravity source was the critical question of the session. Two competing spirals curving from opposite portals toward a central convergence point -- either that produces something spectacular or it produces chaos.

Variation C's dark room images scored 9.15 and 9.45. Not chaos.

The 9.45 image produced an emergent behavior I had not prompted for: where the two spiral fields met at the center of the frame, Firefly generated a lens flare burst -- a visible interference point. The spirals do not just coexist. They collide. Two gravity sources creating a visible intersection in the physics is both scientifically coherent and compositionally elegant. I did not ask for that. The model derived it from the physics logic.

That convergence point became the foundation of every subsequent variation.

Midway through testing, two approaches had emerged with genuinely different strengths.

The spiral variations were producing the highest technical scores. Physics drama, visual complexity, light that looked like it was behaving according to laws. But they were slightly abstract. The images reward study. They do not necessarily stop you in one glance.

Variation E took a different approach: three circular portals arranged horizontally across a wall, a figure standing silhouetted in front of them. No spiral, no physics language. Just a person standing at the intersection of medieval fire, present cityscape, and future rooftop gardens.

That image is immediately legible. You understand the concept in one second. A human at the crossroads of all three times -- past burning to the left, future growing to the right. The emotional impact is direct where the spiral impact is earned through attention.

Neither approach is better. They are different articles. The spiral images answer the question: how does temporal physics look? The silhouette images answer: what does it feel like to stand between your past and your future?

I kept both directions in the session because they produce genuinely different educational content.

The final prompt incorporated every finding from the session at once. Not iterating on one variable. Locking all of them.

Dark atmospheric contemporary loft with exposed brick walls. Gold-rimmed portal frame left, green-white rimmed portal frame right. Dual mathematical spirals toward central convergence point. Left portal: Victorian library with floor-to-ceiling bookshelves, ornate chandelier, mahogany writing desk. Right portal: near-future workspace with curved timber desk, wall of living botanicals, warm recessed panels, single thin display screen. Warm amber prismatic refraction along the left trajectory, soft green-white prismatic refraction along the right. Visible color zones on dark floor. Atmospheric haze revealing competing light rays meeting at center of frame.

Six generations. Average: 9.24. Winner: 9.85.

The portal rim colors are worth explaining because they represent something I had observed throughout the session without making explicit. In earlier variations, Firefly had spontaneously started differentiating the portals architecturally, adding a gold frame to the past portal and a white or silver frame to the future portal without being asked. The model was solving a visual disambiguation problem on its own. When I made it explicit -- "gold-rimmed frame left and green-white rimmed frame right" -- it executed correctly across all six generations.

Converting unprompted emergent behavior into reliable syntax is one of the more satisfying moments in systematic prompt research.

The 9.85 winner produced the infinity figure-8 interpretation: the dual spiral physics resolving into a clean infinity symbol connecting the two portals, with gold light pooling on the dark floor beneath the past portal and green-white light pooling beneath the future portal. Geometrically precise. Conceptually perfect. Past and future connected by infinity.

That image did not require much interpretation. The concept was self-evident.

Full score progression from first test to final synthesis:

That final session average, 9.24 across six images, is higher than the peak single-image score from Variations A, D, or E. Locking every variable simultaneously raised the floor, not just the ceiling. The lowest-scoring final image was 8.70. In Variation A the lowest was 7.00.

The comparison that matters most: the two-era competition entry averaged 88.75 in its final session and peaked at 92.60. The three-era final prompt averaged 92.40 and peaked at 98.50. Adding a third era and a second gravity source did not make the concept harder to execute well. It gave the concept more to work with.

The dark room rule became absolute. I knew from the original session that dark backgrounds outperformed light ones. What I did not know was how reliably every bright contemporary room would fail -- not sometimes, but every time, across 30 images spanning seven prompt variations. This is not a preference. It is a rule.

The future portal is harder to describe than the past. The past has a century of visual reference material. Every historical era has a clear visual vocabulary. The future only has what you give it. "Futuristic" is not a description. Curved timber and a living wall is.

The convergence point instruction does not produce one thing. It opens a vocabulary. "Two gravitational fields meeting at central convergence point creating visible light interference pattern" produced concentric rings, an infinity figure-8, an X-pattern crossing trajectory, a lattice crosshatch, a floor mirror reflection, and an atomic orbital pattern -- across ten different images. The infinity symbol as a physics outcome for two opposing gravity sources is technically accurate and immediately resonant. The model derived it. I just described the physics.

The future portal language is fixed. Do not improvise on it until you have the base concept working. The past era is the only required swap.

For [YOUR PAST ERA], name every object. The Victorian portal should have bookshelves with a specific descriptor (floor-to-ceiling), a light source (ornate chandelier), and a furniture piece (mahogany writing desk). That level of specificity is what separates a portal destination from a glow in the background.

Proven past eras from this session:

Victorian library:

Medieval great hall:

For [YOUR PAST ERA GLOW], match the lighting temperature to the era: "warm golden Victorian" or "amber medieval firelight." This creates the three-color floor gradient that makes the image readable as a timeline.

Eras to avoid: Art Deco and 1920s interiors. Firefly renders them indistinctly from a general "vintage" aesthetic. The era needs to be visually unambiguous.

Expect variance. The convergence point language will produce different physics interpretations across generations. That is not a failure. It is the instruction working as intended. Run a minimum of four generations before evaluating. The 9.85 was generation five of six.

Testing methodology: Firefly Image 5 (@adobefirefly). All images scored using a weighted 5-dimension rubric: Visual Quality (30%), Prompt Alignment (25%), Consistency (15%), Uniqueness (15%), X Engagement Potential (15%). Minimum 4 generations per variation before drawing conclusions. 30 images across 7 variations. Session winner scored 9.85.