The Broken Heart AI photo is a still portrait of you in a rainy night scene next to a large cracked red heart. You upload one photo, pick one of three scenes and get a vertical 2:3 image in about a minute.
The look is also searched as broken heart AI photo editing, broken heart AI prompt or sad AI photo editing. It is not brand new: the earliest dated tutorial we found is from 1 November 2025, and new posts were still appearing on 10 November 2026. This guide goes deeper than the page FAQ, covering the source photo, the scenes, failures, posting and what not to make.
Key takeaways
- The result is one still 2:3 image, not a video, made from a single photo of you.
- Use a sharp, evenly lit photo, full-length for the kneeling and wall scenes.
- Three scenes (kneel, wall, hold) read differently, so try all three.
- Label the post as AI-made, and never use children's or other people's photos without consent.
- Each image is 1 credit, and packs start at $6.90 for 10 credits.
In this guide
- How do I shoot an original photo that works?
- Which of the three scenes fits which mood?
- Why does the heart look normal, or not cracked?
- Why did my face change, and how do I keep it?
- Should I make it in ChatGPT or Gemini myself, or use videosi?
- How do I post it, pick a song and add an AI label?
- Can I use it as a profile picture or cover image?
- What should I not make with it?
- What does it cost, and how many versions should I try?
How do I shoot an original photo that works?
Use a sharp, front-facing photo where your face is clearly visible and the light is even. Window light or open shade beats a harsh lamp, because soft light keeps your face accurate and gives the model clean edges around your hair and shoulders.
Decide the scene before you shoot. The kneeling scene and the neon wall scene show your whole body, so a full-length photo works best for those. If your photo stops at the chest, the model has to invent your outfit and legs, and that is where likeness and clothing drift start. The holding scene is a closer portrait, so a head and shoulders photo is fine there.
Skip sunglasses, hats that hide your hair and heavy beauty filters, since the AI copies what it sees. Keep one person in the frame. The scenes have room for one, so crop a group photo before you upload it.
Which of the three scenes fits which mood?
The three scenes on the Broken Heart AI photo page read differently, so match them to what you want to say.
Kneeling in the rain is the most dramatic. You are on a wet street with roses and a cracked red heart on the ground while a couple walks away, out of focus, behind you. It suits a full "this really hurt" caption and works best from a full-length photo.
The neon wall is the moodiest and the most stylish. You lean on a graffiti wall with your head down under a huge neon broken heart, in red and teal light. It feels like a film still, so it fits a quieter, cooler tone.
Holding the heart is the most intimate. Your eyes are closed, your face is tilted up in the rain, and a glowing cracked heart is held against your chest. It is the easiest to crop for a profile image and the least busy in the background. Since the scenes cost the same, make all three from one photo and post the one you like best.
Why does the heart look normal, or not cracked?
A clearly broken heart is the whole point of the picture, so a plain heart is a real failure. Chat prompts often cause it, because they rarely ask for a glossy object cracked in two with a glowing crack, and the model falls back to a normal heart or a flat drawing.
The videosi setup describes the heart explicitly and says it must always be clearly broken. If one result still comes out with a weak crack, generate again. Each run is a fresh draw, and the heart, the hands and the background can differ from one run to the next.
It also helps to know that the image has no text. The scenes ban captions, logos and watermarks, so anything you want to say belongs in your post, not in the picture.
Why did my face change, and how do I keep it?
When a result looks like a model-like stranger, the cause is usually one of two things. Either the tool never locked identity, or the source photo gave it too little to work with. Chat prompts that only say "make it sad" are the classic case, because nothing tells the model to keep your face.
Here your photo is the only source for your face, hair, age, skin tone, build and clothes, and the setup says not to beautify or slim you. The rain, light, heart and background are the new parts. Likeness is still never guaranteed, so judge the first result honestly.
If the face drifts, retake the photo closer and in better light, remove glasses or a hat, and run it again. A failed or blocked generation returns its credit automatically.
Should I make it in ChatGPT or Gemini myself, or use videosi?
Doing it yourself is a fair way to learn the look, and it is how the trend started: most tutorials are screen recordings of a long prompt pasted into ChatGPT or Gemini after the finished image. The earliest dated tutorial we found is a TikTok from 1 November 2025, and tutorials showing 77,000 to 216,000 views when we checked (unconfirmed) followed in March, April and June 2026.
The trade-off is repeatability. Long pasted prompts tend to add extra text, captions and faces, flatten the picture when depth of field and rain are not described, and change your face when identity is not locked. You also have to find, paste and tweak the prompt every time.
The videosi page runs OpenAI's GPT Image 2.6 with the identity lock, scene, look, heart and rules already written. You upload one photo, pick a scene and press Generate, so there is no prompt to paste and the three scenes stay consistent.
How do I post it, pick a song and add an AI label?
The result is a still vertical 2:3 image with no sound, so post it as a photo or carousel, or put it into a short reel with your own sad song. The trend uses many different sounds, and there is no single required track. Use music you have the right to use.
Label it. TikTok's synthetic media policy asks creators to label realistic AI-generated images, audio or video, and says a label, sticker or caption can meet that, as TikTok describes in its newsroom post on AI labels. Meta says it adds "AI info" labels on Instagram and Facebook when it detects AI signals or when people disclose, and expects people to disclose AI-generated uploads, in its approach to labeling AI content.
A plain line such as "made with AI" in the caption is enough for readers, and it protects you if the platform adds its own label.
Can I use it as a profile picture or cover image?
Yes, with one note on shape. The image is vertical 2:3, so a profile circle will crop it. The holding scene crops best, because your face and the glowing heart sit near the centre. The kneeling scene has the most going on at the bottom of the frame, so it suits a story or a cover better than a small avatar.
Check the crop before you set it. Put the picture in your platform's editor and make sure your face and the cracked heart both survive the circle. If they do not, use the full image as a post and a plain portrait as the avatar.
What should I not make with it?
Keep it to yourself and to people who agreed. Photos of children are not allowed, and no image is generated if the photo shows a child. The scenes are built for one adult only.
Do not upload an ex's photo to make a "revenge" picture, and do not use anyone's photo without their consent. A heartbreak portrait with a real person's face can read as an accusation or a joke at their expense, and it can break platform rules on impersonation and harassment. The setup also bans blood, so the look stays sad rather than graphic.
If you are really struggling after a breakup, a picture will not fix that. Talk to someone you trust or a local support line.
What does it cost, and how many versions should I try?
Each image is 1 credit, and packs start at $6.90 for 10 credits. A good first session is three images, one per scene, from the same photo. That shows which scene your photo suits before you spend more.
If a result has a flawed heart or hand, generate again rather than editing around it. A failed or blocked generation returns its credit. Ready to try it? Open the Broken Heart AI photo page, upload one photo and pick a scene.
