I'm Sorry Miss Meadow, These Vids Ain’t Forreal
An Experiment in Image-to-Video AI, Starring a Deeply Confused Dog
I’ve been messing with image-to-video generation recently. It’s fascinating, sometimes frustrating, and often deeply funny.
The results? Flawed but promising. Generative video still feels like the early days of Vine—limited in length, full of promise, and disproportionately funny when dogs are involved.
For this experiment, I used Adobe Firefly’s Image-to-Video tool1 to animate my dog, Meadow, across five wildly different scenarios.
Let’s walk through each.
Eleven from Stranger Things
Original Image:
Generated Video:
Thoughts:
I thought this would be a bit better. Meadow stands in a retro kitchen, waffles floating around her in perfect orbit. Her eyes are locked in. Her sweater is cable-knit. She is… channeling something.
I had mentioned Stranger Things in the prompt which resulted in it being changed during generation due to Adobe’s content policy. Turns out, models often ignore or filter pop culture names to avoid copyright risks. I think this resulted in the video being slightly different than I expected, but it’s still pretty good.
Meadow as a Hibachi Chef
Original Image:
Generated Video:
This was the one that disappointed me the most.
The problem might’ve started with the image itself: I used a version where the onion volcano was already mid-eruption. Firefly kinda just... kept it there. The animation felt weird and unnatural. Next time, I’d use an image where she’s about to stack the onions—give the model something to do, not just something to extend.
Lesson learned: more nuanced = more fragile. These tools are still figuring out how to make subtle actions look real instead of like a school play directed by DALL·E.
CPR Gym Teacher (or Close Enough)
Original Image:
Generated Video:
This one hurt a little. The idea was simple: Meadow as a gym teacher demonstrating CPR on a dummy. Whistle around her neck. Serious look. Classic school gym.
Instead, we got a dummy that received no chest compressions and a final frame where Meadow has… two sets of legs?
In theory, this should’ve been easy. AI is cool, but it's clearly not magic. Yet.
Meadow as the Janitor from Good Will Hunting
Original Image:
Generated Video:
Being a Frenchie in Boston, I had to see if Meadow could pull off Good Will Hunting energy. The original image? A+ meme. The video? Eh.
Firefly didn’t know what to do with the equations. Instead of writing convincingly, Meadow kinda hovered her paw near math symbols and new ones popped up.
She also devolved into a less photorealistic version of herself at the end. Honestly? Still pretty proud. She may not be solving for x, but she looks like she’s thinking really hard about it.
Meadow on the Joe Rogan Experience
Original Image:
Generated Video:
This one turned out the best—probably because it was the simplest.
There’s not much going on in the image besides Meadow sitting like a canine Musk and puffing a doobie. That clarity helped Firefly shine. The smoke effect looked decent although she takes a much bigger hit than I anticipated.
Final Thoughts
I tried five. I’m not amazed. But I’m not mad either.
Generative video still has a long way to go. It can’t handle text. It gets confused by complex movement. And if you ask it to do anything subtle or realistic, you’d better prompt like you’re directing Oppenheimer.
That said, the results are funny, occasionally brilliant, and deeply absurd.
I’m sure these models will get better. Maybe soon, Meadow will be able to star in a full-length Air Bud Multiverse feature. But for now she’s just a math-solving, onion-flipping podcast queen.
—Charlie, Digital director of Dog Does Stuff
Appendix
I explored a few different AI video generation tools, mostly through recommendations and quick searches via ChatGPT and Perplexity.ai. While I ultimately landed on Adobe Firefly for reasons I’ll explain below, I thought it might be helpful to document everything I tested (or tried to test) along the way.
Adobe Firefly
The one I actually used. (And paid for.)
You get two free 5-second generations without needing to enter credit card info, and $9.99/month unlocks 20 more clips. No watermark, high resolution, and I trust Adobe.
Pixverse.ai
Three free generations, but all were 360p and included a watermark.
Honestly, it worked fine—comparable to Firefly in output—but I leaned toward Firefly simply because it felt like a safer bet.
Runway Gen-2
This one kept popping up as the “best in class,” but I couldn’t test it. The free trial didn’t give me enough credits to generate a full video—and I wasn’t ready to pay before seeing results.
Leonardo.ai
A well-known image generation tool with growing video capabilities. Unfortunately, their image-to-video tool costs 250 credits per render, and I only had 150 remaining after my image generation test for the previous post so I didn’t try it.
The Prompts
These are the image-to-video prompts I used in Adobe Firefly.
Eleven from Stranger Things
Create a 5-second cinematic video animation from this photorealistic image of a French Bulldog in a sweater standing in a retro kitchen. Add supernatural movement to five floating waffles around her, rising slowly and pulsing as if controlled by telekinesis. Her eyes should glow faintly, and the lighting should flicker like from an old fluorescent bulb. The atmosphere should feel tense and cinematic, like a moment from Stranger Things.Hibachi Chef
A French bulldog wearing a white hibachi chef’s uniform and red hat stands behind a teppanyaki grill. The dog holds two spatulas and performs the onion volcano trick—stacking onion rings into a tower and igniting them to create a tall flame. Animate spatula flipping, flames bursting from the onion tower, and reactions from a group of laughing diners in the background. Warm restaurant lighting, photorealistic.CPR Gym Teacher
A photorealistic 5-second training video set in a high-school gymnasium with a french bulldog performing CPR on a CPR dummy.Janitor from Good Will Hunting
A French bulldog standing upright in khakis and a navy polo shirt faces a chalkboard covered in complex equations in a dimly lit school hallway. The dog raises one paw to write on the board with chalk while glancing at the equations thoughtfully. Animate subtle paw movement as chalk writes and erases, small particles of chalk dust falling, and slight camera movement that focuses from chalkboard to the dog’s intense gaze. Academic lighting, photorealistic.Guest on The Joe Rogan Experience
A French bulldog sitting upright at a podcast table across from a bald man with headphones, in a red studio inspired by the Joe Rogan Experience. The dog calmly holds a lit joint in one paw and exhales small curls of smoke. Animate subtle head tilts and a relaxed, thoughtful posture as if responding to a question. Include a gentle camera zoom and ambient smoke drifting upward. Warm studio lighting, photorealistic.Firefly’s Image-to-Video generator is still in Beta.
I liked the Joe Rogan - but if you swapped him for literally anyone else. :) When do you find the time??