Scarlet Pictures
All journal entries

A Face Only A Mother Could Love Review: The Craft Behind a Celebrated AI Short

  • AI
  • Film review

A Face Only A Mother Could Love is a 7:50 short written and produced by Robert Gaudette, and it has had a remarkable year. It won at Runway’s AI Festival in June, then took first place at the Reply AI Film Festival in Venice in September, where it was chosen from more than 3,000 submissions from 76 countries. You can watch it on Gaudette’s YouTube channel, where it has been up since April.

The story is simple and big-hearted. Marcel Dupont is 38, lives alone in Paris, and has a face the world tends to move away from. His late mother told him he was beautiful. Every evening he dances alone in his kitchen with one hand raised for a partner who isn’t there. Then, on Halloween night, a masked woman sits down beside him on a park bench and does not leave.

For anyone making AI films, the most rewarding question to ask about this short is why it works so well. The answer is mostly in the writing, and in how thoughtfully the script is built around what generative video does best. Gaudette made a series of smart choices, and each one is worth studying.

The narrator does the acting

Almost the whole film is carried by a third-person narrator with a dry, precise delivery: “He takes his coffee black. He’s watching his weight.” The narrator reports Marcel’s view of the world without correcting it. The baker has never learned Marcel’s name, and “Marcel interprets this as professional discretion.” His father leaves when Marcel is a child, and “Marcel understood.”

The comedy and the tenderness both live in the gap between what the voice says and what we understand. It is a classic storytelling device, and here it is also a clever production decision. Subtle facial performance is still one of the hardest things to get from AI video, and this film rarely asks a face to carry a beat on its own. The narration supplies the subtext, so the images only need to show a man walking, waving, ironing, or sitting still, and every one of those images lands.

According to the filmmaker’s YouTube description, the narration was made with ElevenLabs voice synthesis and the score with Suno. The few dialogue lines were lip-synced in generation and then redubbed for voice consistency. That extra pass shows. The voice is the performance, and it was tuned like one.

The lesson: a well-written voice can carry a film’s emotion beautifully. Write it like a lead role.

Specific numbers do the characterization

The script keeps counting. Marcel owns 41 masks. He has danced alone for 11 years. The woman sits down “at 9:47 on the 31st of October.” They talk for three hours and 40 minutes, and “Marcel counted.”

Each number tells you something about him: that he is careful, that he keeps records, that this conversation is the event of his life. It is lovely, economical characterization, and none of it costs anything to render. That is exactly the kind of trade AI filmmakers can learn from. Put the specificity in the words, where it is cheap and fully under your control, and let the images stay clean and simple.

A story about faces, told through masks

The central motif is masks. Over seven years his father gives him 41 of them, and once, with nothing else on hand, makes one from a paper bag “in three cuts.” On Halloween both Marcel and the woman wear gold half masks. A film about how a face is seen is told largely through objects that cover faces, which is an elegant piece of thematic design.

It may also have helped with production, whether or not that was the plan. A mask is a rigid designed object, which is easier to keep the same from shot to shot than a moving human face. Marcel’s own face is a stylized caricature, with a long nose and heavy features, rather than an attempt at photorealism. That gives him a memorable, expressive look and gives the model a strong design to hold. Gaudette’s description says character consistency came from generating character sheets first, and the result is impressive: Marcel is instantly recognizable as a boy and as a man.

Inserts carry the plot

Look at where the story turns and you will usually find an object, not a face: a needle dropping onto a record, scissors cutting a paper bag, an alarm clock at midnight, an iron going over a shirt (he irons it four times), old boots being polished, a pigeon taking the last of the bread.

Inserts like these play to the strengths of generated video: one subject, one action, no dialogue, a few seconds long. Classic short-film grammar already leans on them, and this film uses them with real confidence. When the emotional peak arrives, the camera is often on hands and feet, and the moment is all the more touching for it.

Short shots, and a ritual that repeats

A scene-detection pass over the YouTube upload finds roughly 150 cuts in 7:50, an average shot of a little over three seconds. Treat that as approximate, since automated detection misses some dissolves and counts some flashes. It is a brisk, lively cut that keeps the story moving, and it keeps each shot short enough that a generated clip stays consistent.

The structure uses repetition just as skillfully. (Light spoilers from here.) Early on, the narrator describes Marcel’s nightly ritual step by step: he clears the kitchen floor, places the needle on the record, raises his left hand to shoulder height, and dances. The film ends on the same words over the same kitchen, with one change in the frame. It is a wonderful payoff. Reusing a set, a camera setup, and a line of narration is efficient, and the emotion comes from the variation, which is exactly how a well-planned live-action short makes the most of a small budget.

Ideas for building on it

A film this accomplished invites thinking about where the approach could go next. A few ideas that could make work like this land even harder:

  • Let a key memory breathe. The memory sequence, where Marcel remembers the boys on Rue Descartes and the names they called him, sits in the fastest-cut stretch of the film. As tools for held facial performance keep improving, giving a moment like this one longer shot on a face could make an already moving scene even more powerful.
  • Let images and narration trade off. The narration is the film’s great strength. In a future piece, an occasional image that adds a new layer beyond what the voice has just said could give the audience even more to discover.
  • Play with the style contrast. Marcel is a caricature, while the woman is rendered close to photoreal. That reads naturally as part of the story, since the whole film is about how differently he is seen. Leaning into the contrast even more deliberately, or bridging it in the frames they share, is a rich creative option for the next project.

What to take from it if you’re making AI films

  • Write to the medium’s strengths. Narration, inserts, and short shots are where current tools are dependable. Design the script around them before you start prompting.
  • Put specificity in the words. Exact numbers, names, and habits cost nothing to generate and do a lot of character work.
  • Choose designs a model can hold. Masks, costumes, and stylized faces stay consistent more easily than naturalistic ones. Build character sheets before you generate shots.
  • Plan repetition. A ritual that comes back with one change is cheap to produce and lands emotionally.
  • Lead with story. The festival audiences and juries rewarded this film for being a film first. That is the bar, and Gaudette cleared it with style.

For context on how quickly things have moved, compare this solo short with the 50-artist collaboration behind Our T2 Remake in 2024, or with the AI video experiments on this site. Two years ago the question was whether AI video could hold a story together for a feature. Now a single writer-producer can win on craft, and A Face Only A Mother Could Love shows how.

Sources

Scarlet Pictures’ own short films are on the films page.