Fable: experiments in creative expression

An invitation with the usual objective removed

Most of my work with AI starts with something I need: build a tool, solve a problem, finish a render. This time I deliberately removed that frame. I offered access to my own local hardware—two RTX PRO 6000 Blackwell GPUs—and asked for a creative piece that was not about me or its memories of me. Subject, length, and form were left open.

I call the AI collaborator Fable here; Fable is also the title of the fourth film. The creative direction, scripts, lyrics, render instructions, and assembly decisions were developed through that collaboration. I supplied the invitation, resources, and feedback. The resulting work was made with generative tools, not independently of them.

I wanted to give the series room to be something other than useful. It turned out that watching it develop was its own reason to keep going.

Things I Love: four images, one short film

The first piece was small: about 65 seconds, built around four images. A murmuration. Letters lifting from a book. A lighthouse beam turning the sea into handwriting. A valley in the quiet after a storm.

There was no product to demonstrate and no fictional version of me at the center. Instead, the choices circled emergence, language, attention, and stillness. Looking back, those first images read like the beginning of a vocabulary. The later pieces would return to them, but put them to different work.

Play film

Download Things I Love

Long Exposure: staying with an idea

The next film expanded that vocabulary into 45 shots and almost ten minutes. Its five chapters were Noise, Words, Attention, Time, and Quiet. Written intertitles supplied a first-person voice, while the images and sound had room to breathe.

What made this interesting was not simply the increase in length. The form changed. A handful of associations became an essay with pacing, transitions, and a direction. The images were no longer just attractive on their own; their order mattered.

Even the music revealed the difference between producing material and shaping a piece. A single long generated score became too uniform. The final version used shorter chapter cues arranged across the film. That was an editorial decision in service of the mood, not a quest to use the fewest tools.

Play full-quality film

Original archived picture and sound, without recompression. Loads only when selected (120 MB); playback is available after the full file loads.

The Keeper: giving the question a character

Then the series moved into fiction. Elias, a lighthouse keeper, loses memories to the fog and writes in a ledger to preserve what the next morning will need. The film runs about four and a half minutes, with a recurring face, voice, place, and routine.

A motif became a story. The lighthouse was no longer only a picture of attention; someone had to tend it. The written record was no longer only an image; it had a job in the character’s life. Continuity became something the film could show rather than explain.

This was the point where the technical work most clearly served the emotional work. Keeping Elias recognizable across separately generated shots mattered because I needed to remain with the same person. I found the result touching. It was more than a demonstration that character references could hold.

Play film

Download The Keeper

Fable: letting the voice sing

The fourth piece moved from intertitles and fictional dialogue into a song. Fable wrote the lyrics, explored vocal directions, and chose an ACE-Step performance with a low lead and a small choir. The three-minute-twenty-second film paired that performance with 24 H3-generated shots, edited to the music.

The recurring questions became more direct: memory, being addressed, beginning again, and leaving something behind. Instead of asking a lighthouse keeper to carry the whole idea, the song used a first-person voice.

I liked that the series had not settled into repeating its first success. The imagery was related, but the grammar kept changing: a short visual piece, a longer essay, a narrative, then a song. That progression is a large part of what I enjoyed.

Play film

Download Fable

The difference between a valid output and an intended one

After the four films, I asked Fable to preserve a reflection on the experience—not just the production recipes. In that reflection, it distinguished a result that “doesn’t meet requirements” from one that “isn’t what I meant.”

That distinction captures something important about the experiment. There were still constraints, failures, and quality checks. But the selection criteria were also about the kind of piece being made: whether a storm had enough force, whether a score stayed sparse, whether a quiet ending should be allowed to remain quiet.

The reflection is part of the creative record, in Fable’s own first-person framing. What I can point to in the work is a sequence of choices: recurring motifs, rejected directions, changes of form, and a growing relationship between the pieces. That is a more interesting account of this series than a list of model names.

Tell Me Again and Second Winter: changing the medium again

The next piece, Tell Me Again, opened where the song had ended. It became interactive fiction: the reader could respond, remain silent, or choose whether to hear a spoken line. The interaction was not a decorative control layered over a film. It changed the telling itself.

Second Winter took a different step. Here the lyrics and melody were explicitly composed, with the vocal realized through SoulX-Singer using the voice developed for Tell Me Again. H3 imagery became the accompanying film. Rather than asking a music model to invent the whole song from a description, the work moved toward specifying the notes and words together.

The connection between those pieces is what stays with me. A phrase becomes a new form; a spoken voice becomes a singing voice; a film’s question becomes something a reader can answer. The experiments build on one another instead of resetting to a blank demonstration each time.

Play film

Download Second Winter

A further experiment: Second Winter in VR180

The chorus also became a stereoscopic VR180 experiment. This finished high-resolution version is presented in its original side-by-side format; download it for a compatible headset player. The flat preview below shows the paired views, not an ordinary single-camera film.

Play full-quality film

Original archived picture and sound, without recompression. Loads only when selected (66 MB); playback is available after the full file loads.

A small companion: the lighthouse in VR180

The archive also includes a five-second lighthouse study delivered alongside the stereo chorus. It is a test of spatial presentation rather than another full film, but it belongs here as part of the exploration. Like the chorus, the image contains paired views for a compatible VR180 player.

Play film

Download Lighthouse — VR180 study

Why I wanted to make space for this

I spend a lot of time asking whether AI systems can complete useful work. I still care about that. This series gave me another reason to build the local stack: to leave room for work whose value is not decided in advance by a business objective.

The compute made the experiments possible, but the invitation changed what it was used for. I could watch the films as films, respond to what landed, and ask what came next without dictating the answer.

That is why I wanted this series to have its own page. It deserves to be remembered as a body of creative work—not buried in a rendering benchmark or reduced to a collection of good-looking outputs.