Production Notes

I finally exported the Bakunawa video for Facebook just after midnight. It was one of those days where I kept telling myself, “Just one more revision,” until I realized the day was already over.

Before I could animate anything, I had to lock down Bakunawa's design. Consistency is one of the hardest parts of AI filmmaking, so I always start by creating a definitive version of the character. I spent time refining the shape of its face, the color and texture of its scales, its eyes, its horns, and its overall silhouette. Once I was satisfied with the design, I could generate new scenes knowing the creature would remain recognizable throughout the video.

The voice-over took much longer than expected.

My original plan was to use the Eleven Multilingual V2 text-to-speech model from ElevenLabs, but getting natural-sounding Tagalog narration from it wasn't easy. Unlike English, Tagalog often requires careful placement of diacritical marks to guide pronunciation, rhythm, and emphasis. I spent quite a bit of time experimenting with different versions of the script, but I still wasn't completely happy with the results.

Instead, I recorded the narration myself and processed it using ElevenLabs Voice Changer. That ended up being the better solution. My raw recording had inconsistent volume, with some lines louder than others.

After running it through the voice changer, everything sounded balanced, clean, and much closer to something you'd hear in a professional recording studio. More importantly, it still sounded like a real person telling the story instead of a synthetic voice.

Even with the Bakunawa video taking most of my attention, I managed to make progress on another important part of the project: Alunsina.

One of the creative decisions I've made for this series is to gradually move away from characters that resemble existing celebrities. While early concept art borrowed inspiration from familiar faces, I want the final cast to have completely original identities. Today, I redesigned Alunsina from the ground up, giving her a face that belongs entirely to this universe.

To establish her appearance, I wrote a detailed prompt describing a young Filipino woman with realistic skin texture, natural facial features, expressive eyes, and soft studio lighting. My goal wasn't to create someone who looked flawless. I wanted someone who looked believable. The result surprised me. The portrait feels so realistic that she looks less like an AI-generated character and more like someone you could actually meet in everyday life.

So far, I've completed her base face along with several camera angles. Having multiple perspectives is critical because every future shot, animation, and facial expression depends on maintaining that same identity.

I also started developing Alunsina's superhero form. The biggest challenge, once again, is the helmet.

Unlike a simple costume change, the helmet has to match the character's face from exactly the same perspective. That means I first need to generate the helmet from multiple angles before I can convincingly integrate it into each shot. It's a slow process, but it's one of those details that makes the difference between a believable character and one that immediately looks AI-generated.

Before calling it a day, I also added another preset to my Cinematic Scene Generator. Every preset I build saves time later by standardizing the project's visual language. Instead of rewriting the same stylistic instructions over and over, I can focus on directing the scene itself. It's a small improvement to the workflow, but dozens of these small improvements are gradually becoming the foundation of how this series is being made.

Some days end with a finished scene. Other days end with a better workflow. Today was a little of both, and that's progress I'll gladly take.

Today's focus was refining my color grading workflow. Instead of treating color as something to adjust at the end of the process, I wanted to build a system that would keep every scene's look as consistent as possible. To accomplish that, I developed a custom workflow that helps maintain a unified color grade across the frames I generate.

Color grading plays a much bigger role than simply making an image look attractive. It helps establish the mood of a scene, reinforces the emotional tone of a story, and gives the entire project a cohesive visual identity. A subtle shift in color can completely change how an audience experiences a moment, whether it's creating a sense of warmth, tension, mystery, or hope.

Gritty War Epic Preset

To make the prompting process more reliable, I also created a collection of color grading presets. Rather than rewriting the same visual instructions every time I generate a new frame, I can now apply these presets to achieve a consistent cinematic look while still making adjustments for individual scenes when needed.

Building these presets should also make future production much more efficient. Instead of starting from scratch with every new image, I'll have a solid visual foundation that keeps the lighting, color palette, and overall atmosphere aligned throughout the series. It's another small step toward streamlining the entire production pipeline and ensuring that each scene feels like it belongs in the same cinematic world.

One of the biggest challenges of using AI tools for 3D model sculpting is maintaining consistency. To achieve reliable results, prompts need to be extremely clear and detailed. Even a small change in wording can affect the final output, resulting in differences in color, lighting, texture, proportions, or overall design from one generation to the next.

There's a common misconception that AI makes it easy to create a short film with just a few clicks. In reality, the process is far more involved. Most of the time, the first image generation is only a starting point. Rarely does it produce the exact visual style you're looking for right away. Instead, the work becomes an ongoing cycle of experimentation, evaluation, and refinement.

A significant portion of the production process is spent rewriting prompts, adjusting design elements, generating new variations, and comparing results against the creative vision. The goal isn't simply to create an image that looks good. The goal is to create images that remain visually consistent across an entire project, especially when multiple scenes, camera angles, and character appearances need to match one another.

The helmet design for Bronze Valkyrie is a perfect example of this challenge. What appears to be a relatively simple piece of armor required more than twenty revisions before I arrived at a design that truly matched what I had envisioned. Each version introduced new ideas, but also new problems to solve. Some designs had the right shape but the wrong materials. Others captured the overall silhouette but missed important details in the ornamentation or proportions.

While the revision process can be time-consuming, it's also one of the most rewarding parts of the workflow. Every iteration brings the design closer to its final form. In many ways, creating with AI feels less like pressing a button and more like collaborating with a highly talented but unpredictable artist—one that requires patience, precision, and a great deal of creative direction to achieve the desired result.

As the project has evolved, the overall direction of the series has become much more defined. The story will follow three central heroines: Midnight Blue Valkyrie, Bronze Valkyrie, and Hunter Green Valkyrie. Although they exist within the same universe, my goal is for each character to have her own distinct visual identity, personality, fighting style, and transformation sequence. I want audiences to recognize each Valkyrie the moment she appears on screen.

One of the biggest creative challenges has been designing the transformation scenes. For many fans, a superhero's transformation is just as memorable as the action itself. It's the moment that reveals the character's true power, so I wanted every sequence to feel like a major cinematic event instead of a simple visual effect. The transformation needed to build anticipation, convey emotion, and end with a dramatic reveal worthy of the character.

Over the past several days, I've been experimenting with different AI workflows and refining countless prompts to achieve that effect. Every detail matters, including the camera movement, the rhythm of the animation, the lighting, the flow of the energy effects, and even the way the costume materializes around the character. Small prompt adjustments can produce dramatically different results, so much of the process has been a cycle of testing, reviewing, and refining until each scene feels polished.

At the moment, Bronze Valkyrie's transformation is the one I'm most satisfied with. Even so, it's still not considered final. The character herself is continuing to evolve, and even her current name is only a working title. That's one of the reasons I enjoy creating these teasers. They give me the freedom to explore new ideas, gather feedback, and refine the creative direction before committing to the final version of the series.

For Bronze Valkyrie's appearance, I took inspiration from actress Sanya Lopez, who has long been one of the most popular fan choices to portray Darna. Many fans have imagined her in the role over the years, so giving Bronze Valkyrie a subtle resemblance to her felt like a fun tribute to that enthusiasm. It's a small piece of fan service, but the character is still being developed as an original heroine with her own story, personality, and place within this universe.

Designing the teaser turned out to be more challenging than I expected, especially when it came to blending the characters seamlessly with their environments. I spent several hours writing and refining prompts, then generating and iterating on images using large language models (LLMs) until I achieved the visual style I was aiming for.

My first concept reimagined Marian Rivera's Darna as a warrior from the Roman Empire. I wanted to create the feeling of an epic one-on-one battle against Medusa, giving the character a completely different historical and mythological setting while still preserving her heroic presence.

After that, I changed the costume from its traditional red to blue. This wasn't just a creative choice—it also made it much easier to use the design as a base for experimenting with different character identities and face replacements during the concept development process.

The story I'm developing centers around three female superheroes, each distinguished by the color of her armor: a brown-armored warrior, a blue-armored warrior, and a green-armored warrior. Each character will eventually have her own unique identity, personality, and role within the story.

The most difficult part of the entire process was transferring facial identities onto the characters. Current LLMs have become much stricter about generating images that resemble real people, particularly when face replacement techniques resemble deepfakes. Because of these limitations, achieving consistent and accurate character appearances required a great deal of experimentation and prompt refinement.

For that reason, the characters featured in this teaser should be viewed as early visual concepts rather than their final designs. Their appearances will continue to evolve, and they will likely look different by the time the full series is produced.

It all started with a simple idea: “Imagine Jewel in the Palace... but make it Pinoy.”

The response was incredibly encouraging, and it made me want to keep exploring the concept. At the same time, I noticed that many Filipinos had grown tired of localized adaptations of South Korean dramas.

That sparked a different question: What if, instead of adapting another foreign series, we reimagined Darna for a modern audience? What if we gave her a cinematic treatment using AI?

That idea became the first teaser.

It was also my first time working with AI tools, so every step of the process became a learning experience. There was a lot of trial and error, and I quickly realized just how much there was to learn.

Darna, in particular, has been one of the biggest challenges because of the fast-paced action. Creating convincing action sequences with AI takes a lot of careful prompt writing, and every scene has helped me refine that skill a little more.