I wanted a werewolf transformation worthy of a horror film

I'm a horror fan. When I set out to make a werewolf Momento, I had films like An American Werewolf in London in mind: the kind of transformation that makes you wince as a face stretches, teeth push through, and a creature takes shape before your eyes. I wanted to bring that feeling to the werewolf Momento.  I wanted our customers to recognize their photo as the change began, then watch their face grow into something terrifying.

When I was making this, I hadn't thought to search for images from the film. I was working from memory of watching American Werewolf in London, a film I hadn't seen in over thirty years. The fact that its transformation had stayed with me for so long was how I knew it would be a great Momento, because its cinematic quality has an impact. 

That's the idea behind MagicLens Momentos: high-quality cinematic digital keepsakes that turn your photos into experiences. For this one, a frightening final picture was only the beginning. I wanted the film of you becoming the werewolf to be worth watching again.

I started with a few one-shot prompts, asking the video model to make the transformation scary. The results looked fake to me. They played like transitions between a person and a monster, with the important part (a gory transformation) missing. 

Image and video generation is probabilistic. A prompt gives the model direction, but it leaves plenty of room for the model to decide how the scene unfolds. I could keep trying and hope for an exceptional take, or work out how to build the effect I actually wanted. Getting this one right took many hours of iterations.  Our tools made iteration possible, and with each step I learned what worked and what didn't; the learning built in, and the final product locked in as a repeatable workflow.  

The process leverages different models for different steps, which I also tested to get the highest quality at the best prices.  This is an important aspect of our Momentos: we build them to the highest cinematic quality while also finding the best price.  

Building the transformation a piece at a time

I had a few werewolf options I considered building from, but none were exceptional, so I decided to work from the selfie through the transformation.  

I slowed the sequence down and added an image of the transformation halfway through. The opening improved, but the rest still rushed. The nose, teeth, hair and body seemed to take turns. I wanted them to overlap, as they would in the scene I was imagining: the snout and teeth growing while the neck and shoulders swelled, hair covering the human ears before wolf ears emerged.

Building separate keyframes gave me control over those moments. Each was a still image at a particular stage; the video model animated the change between them. I could inspect a weak stage, correct it, and try that part again.

That's what I value about Studio Pro: I can work incrementally, keeping the prompts and workflow steps explicit and saving the versions I want to build on. Each change has a place in the sequence, rather than depending on a long conversation to remember every earlier decision. It gives me control over the process; the generated images and video still need review.

Keeping the subject in the shot

As the creature grew, another problem appeared. The head became smaller in the frame and more of the body came into view. It looked as though the camera had stepped backwards to make room.

The end image had a wider crop, so the video was moving towards that composition. I rebuilt the keyframes around the original photo: the same head position, the same head size and the same amount of body in view. The shoulders could swell and the jacket could tear within that frame. The camera had to stay with you.

A second generated keyframe used as a reference could also pull the composition away from the previous frame. For this sequence, I edited each keyframe from its predecessor and repeated the framing instruction. The resulting twelve-second test held its composition. That gave me a specific requirement to check in every later take.

Making the change feel physical

The middle needed more of the detail that made me want to build this in the first place: stretched skin, raised veins, crowded teeth. The instructions had to describe those changes happening. Describing a finished werewolf left the model too much freedom to reveal one all at once.

Once the movement improved, I could see a pause between clips. A frame-by-frame check found roughly 0.9 seconds of almost still footage at the end of the first clip. Trimming it removed the hesitation without paying for another generation.

The ending took several attempts. One creature was too furry. Another settled into a calm face with its mouth closed. I made a leaner final image, kept the mouth open in a roar and added a keyframe so the last change had room to develop.

The version I finally approved was fourteen seconds long: four clips, continuous growth, the head held in place and the roar carrying through to the last frame. I now had a film I could use as the reference for the experience.

Returning to one shot with a storyboard

With the sequence worked out through my multi-render workflow, I now needed to bring it all together in a one-shot workflow.  Don't get this confused with a one-shot prompt, because it's far from that (the Studio Pro diagram below shows the staged workflow I developed first). 

 The one-shot is one video render with one soundtrack from start to finish.  Jason perfected the technique on the Tooth Fairy, our first-ever Momento.   It lets us control the video model's output in a variety of scenarios, knowing we'll have to deal with a wide range of images uploaded by users.  

At this stage, I moved from Gemini to MiniMax H3 as the cheaper option for the one-shot tests. With the sequence worked out, I could compare its result against the staged reference and judge whether the lower cost preserved the effect.  Overall, while the effects were slightly more animated, the storyboard was preserved, and the Momento still maintained its high-quality cinematic experience.