The short version: a good ChatGPT prompt for Ghibli style art names the medium (hand painted animation cel), the light (soft, golden, diffused), the specific film reference (Spirited Away, My Neighbour Totoro, or Kiki’s Delivery Service), and tells the model exactly what to keep from your original photo. Skip the word “anime,” it pulls you toward something sharper and more generic than what you want.
Why most people’s Ghibli prompts come out flat
I watched this trend explode in spring 2025 when OpenAI rolled GPT-4o’s native image generation into ChatGPT and half of Twitter turned their family photos into Studio Ghibli frames within about 48 hours. Sam Altman posted that the servers were “melting” from demand. I tried it myself that first week, typed “make this Ghibli style” into ChatGPT, uploaded a photo of me on a client call, and got back something that looked like a mid budget anime, not a Miyazaki frame. Flat colours, no warmth, wrong eyes.
The problem is that “Ghibli style” on its own is doing almost no work. ChatGPT has seen millions of anime images and a much smaller, more specific slice of actual Studio Ghibli frames, hand painted backgrounds, that particular soft haze of light through trees, the slightly wobbly line work that comes from cel animation rather than digital vector art. If you don’t name those things, the model defaults to the bigger, blander category.
The exact prompt structure that works
After maybe thirty tries across different photos, this is the structure I settled on, and it is the one I now teach clients when we talk through using ChatGPT for trending photo edits on Instagram:
- Name the exact reference film, not just “Ghibli,” because Spirited Away and My Neighbour Totoro have different palettes
- Describe the light source specifically: “soft afternoon light,” “golden hour glow,” “overcast diffused light”
- Tell it what to preserve: face shape, hairstyle, clothing, pose, background objects
- Ask for hand painted texture and soft cel edges rather than sharp digital lines
- Specify the mood word: whimsical, melancholic, cosy, adventurous
My actual working prompt, the one I paste into ChatGPT after uploading a photo, reads like this:
“Reimagine this photo in the hand painted animation style of Studio Ghibli films like Spirited Away and Kiki’s Delivery Service. Keep the person’s face, hairstyle, and pose exactly as they are in the original. Use soft, warm, slightly diffused lighting like golden hour. Add gentle cel animation texture with soft rather than sharp edges, and a warm, whimsical mood. Keep the background setting recognisable but painted rather than photographic.”
That last line matters more than people think. Without it, ChatGPT will sometimes swap your kitchen for a generic Japanese countryside because that’s what “Ghibli” statistically correlates with in its training data.
Step by step, in plain terms
- Open ChatGPT (GPT-4o or later) and upload a clear, well lit photo, ideally one where the face isn’t turned too far from camera
- Paste a prompt that names the specific film, the light, and what to preserve, using the structure above
- Wait roughly 20 to 40 seconds for the first render
- Ask for one specific change rather than starting over, for example “make the light warmer” or “soften the edges more”
- Download the result at full resolution before you do anything else, because ChatGPT sometimes compresses images shared back through the chat window
I’ve found two or three rounds of small edits gets a far better result than trying to write the perfect prompt first time. Treat it like art directing a very fast, very literal illustrator rather than issuing a single command.
What I did with it
I ran this for a client, a small events business in Bristol, who wanted a set of six team photos turned into a Ghibli style “meet the team” page for a spring campaign. We tested the generic prompt first, “Ghibli style version of this photo,” and got results that were unusable, three of the six had subtly wrong facial features, one person’s glasses vanished entirely. Once we switched to the fuller prompt naming the film, the light, and what to preserve, five of six came back close enough to use with only minor touch ups in Canva afterwards. That’s not a perfect hit rate, but it’s the difference between a campaign you can ship and one you quietly bin.
The whole set took about ninety minutes including the touch ups, versus the quote we’d had from an illustrator for something similar, which was £180 per character. For a “just testing the water” campaign, ChatGPT got them 80% of the way there for free.
The bit nobody tells you before you post one of these
Here’s what most posts on this topic skip past. Hayao Miyazaki himself has been on record, in the 2016 NHK documentary about his studio, calling AI generated animation “an insult to life itself,” reacting to a demo of an AI walking figure that his own team showed him. He wasn’t talking about ChatGPT specifically, that tool didn’t exist yet, but the sentiment from the man whose name is on the style you’re recreating is not warm. Studio Ghibli as a company has never publicly endorsed or licensed this use of its aesthetic, and there’s no clean legal ruling yet on whether training a model on a studio’s visual style and then generating lookalike images infringes anything. In the UK and US right now it sits in a unresolved grey area.
What that means practically, for a small business, is not “don’t do it,” it’s “don’t build a paid product or a client deliverable around it without flagging that to the client.” Using it for a fun personal post is low risk. Using it to sell “Ghibli style branding packages” as a service is a different level of exposure, both reputationally, given how public the original artist’s objection is, and potentially legally, once the copyright cases working their way through courts against other AI image generators start producing rulings that could apply here too. I say this to clients directly: enjoy the trend, don’t build a business model on top of someone else’s unresolved legal grey area.
Getting closer to real Ghibli texture, not just the label
A few smaller tweaks made a bigger difference than I expected. Adding “muted, slightly desaturated colour palette” stopped results looking like glossy modern anime. Asking for “painterly clouds with visible brush texture” fixed backgrounds that otherwise looked digitally smooth. And specifying “a slightly larger head to body ratio, like a Ghibli character design” nudged faces toward the right proportions when a straight photo edit made people look too realistic and uncanny sitting next to a painted background.
If you’re doing a batch, for a team page or a set of social posts, keep one “master prompt” as your base and only change the small details, the person’s name and what to preserve, per image. That consistency is what makes a set of images look like they belong to the same campaign rather than six separate experiments.
Where this fits in a marketing plan
I’d treat this as a short lived engagement tool, not a brand identity. Trend driven visual styles like this have a shelf life measured in months, the Ghibli wave was everywhere in April 2025 and had mostly cooled by summer. Use it for a single campaign burst, a launch post, a “behind the scenes reimagined” series, something with a clear start and end date, rather than rebuilding your whole visual identity around a style you don’t control and a founder who has publicly said he dislikes what you’re doing with it.
Related reading: chat gpt portrait.
Frequently asked questions
What is the best ChatGPT prompt for Ghibli style art?
Name the specific film (Spirited Away, My Neighbour Totoro, or Kiki’s Delivery Service), describe the lighting as soft and warm, tell ChatGPT exactly what to preserve from your original photo (face, pose, clothing), and ask for hand painted texture with soft cel edges rather than sharp digital lines. Avoid the word “anime” on its own, it pulls the result toward a more generic style.
Is it legal to make Ghibli style images with ChatGPT?
There’s no settled legal ruling yet on whether generating images in a specific studio’s visual style infringes copyright, and Studio Ghibli has never licensed this use. Personal, non commercial use is low risk. Building a paid product or client service around it carries real reputational and potential legal exposure while the broader AI copyright cases are still working through the courts.
Why do my ChatGPT Ghibli images look more like generic anime?
This usually happens because the prompt only says “Ghibli style” without naming a specific film, light quality, or texture. ChatGPT has seen far more generic anime than true Studio Ghibli frames, so a vague prompt defaults to the bigger, blander category rather than the specific painted look.
How long does it take to get a good Ghibli style result?
Each render takes about 20 to 40 seconds, and most usable results need two or three rounds of small edits after the first attempt, asking for warmer light or softer edges rather than starting the prompt over. Budget 10 to 15 minutes per image if you want a clean result.
