Using GenAI To Make A Visual Personalised Gift
- Harshal
- Oct 8
- 3 min read
Verdict After Experimenting With AI Image and Video Generation Tools
I made a gift for my wife using generated AI tools to edit videos and images. Here, I list out what I tried and the tools that worked or didn't work. I hope this gives you some ideas for GenAI tools and prompts you can use.
I spent 31 minutes writing this and a few hours making the gift. You need 2 minutes to read this.
Google Photos, Without AI
I first went through Google Photos to find one photo for each of the last 15 years, and I wish there were a better way to search for those photos. For example, I was wondering if I could make a n8n workflow that takes a prompt from me and searches through Google Photos using the Google Photos API, but I didn't try that this time.
Ask AI To Make Prompts
After I had those 15 photos, I asked ChatGPT to think of different prompts that are good for a romantic look. Something can be cartoony, watercolor, or something. It's not looking for vector art, it's not looking for abstract, but something which still preserves the resemblance and likeness of the photos to the original, so that it still looks like it's the couple or their family, even if the look and feel is different. But the people should look the same.
Trying Out AI Image Generators
I tried:
Perplexity’s Flux
ChatGPT 5
Google Gemini Nano Banana
Nano Banana was the fastest, and it distorted the images the least when using a watercolor prompt.
ChatGPT 5 was the slowest to generate images, and it distorted images a lot when using the watercolor prompt. It made a fat person look fatter and a slanted smile look crooked. It retained perfect details like the background or text on a t-shirt.
Flux via Perplexity was faster than ChatGPT 5 and its watercolor images had fewer distortions. But, the images didn’t reach my desired outcome.
I eventually used most of ChatGPT 5’s output.
Image-Gen Prompts
Pixar-look:
Convert the given photo into Pixar-style 3D. Keep the subject’s facial features, pose, hairstyle, and clothing recognizable, but stylize everything with smooth 3D Pixar character traits. Use large expressive eyes, soft facial shading, rounded cartoon proportions, and clean skin texture. Apply soft cinematic lighting and a warm color palette. Background should be softly blurred or match Pixar-style scenes (e.g., subtle pastel tones, minimal noise).
Watercolor-look:
Watercolor painting of the uploaded photo with beautiful, realistic facial features, gentle expression, and soft skin tone. Maintain original likeness with elegant proportions. Use light ink outlines and pastel brush strokes on textured watercolor paper. Avoid distortions or exaggeration. The mood should be romantic, dreamy, and timeless — like a hand-painted memory portrait in a high-end art book.
Stitching Images Into Video
I tried or considered:
Google photos
Veed (in the past)
Descript
OpenAI Sora
I have tried Veed and Clideo in the past. I’m a fan of Descript for editing podcasts, and I now have Descript premium, so I tried it. I knew the work involved to do this in Veed, so I was hoping Descript would be easier, but it was not.
I wanted to give Sora photos of my wife from childhood to adulthood and make it create a morphing video of her growing from childhood to adulthood, but it couldn’t do that. It could only generate a video from one photo.
I tried Descript’s AI “Underlord” but it was very slow and erroneous.
Verdicts
Like Descript, there are many products now where the original product was great, but they risk losing product-market fit if they don’t add GenAI. But, adding GenAI as-is results in a slow and bad experience. Adding it well can give a Purple Cow.
My Insights
Next time, I’ll iterate on finding a prompt that works well through whichever AI model will generate the fastest. The speed allows me to iterate to find the right look and try variations. I’m open to trying other video generation tools. Let me know your suggestions to convert a series of photos into a video!
Related:
