By the end of this guide you will have trained a personal AI image model on a set of your own photos, then used it to generate new images of yourself, your product or your art style. The job is done by an image generation service that supports custom model training: you upload a batch of photos, the service fine-tunes a model on them, and you write prompts that use it.
If you want to train an AI on your own images, the main choice is what the model should learn: your face, a product, or an art style. This guide covers each case and compares seven tools from our Image generation category.
Key takeaways
- Collect varied, sharp photos of one subject, upload them to a training tool, wait for the model to finish, then generate test images and adjust.
- The main decision is the subject: selfie tools such as Photo AI and MyPicNow target faces, Astria and Flair AI target products and brand imagery, and Krea, Leonardo.Ai and Playform suit styles and artwork.
- Photo counts vary by tool, from 6 to 10 images for Flair AI up to 30 or more for Playform, so check the requirement before you gather files.
- Only train on photos you own or have permission to use, and never on other people's faces without their consent.
What you need before you start
- A set of photos of one subject. For a person, ideally varied poses, expressions, lighting and backgrounds. For a product, sharp images from several angles. For a style, consistent examples of your own artwork.
- A plan that includes training. Leonardo.Ai offers personal model training to subscribers, MyPicNow says it has no free version, and Playform charges credits for training on every plan.
- Time. Gathering photos takes under an hour. Training time varies: MyPicNow says roughly 30 minutes on average, while Photo AI says lower plans can take 30 minutes to 2 hours.
- Difficulty. Easy for the selfie and headshot tools, moderate for API-based workflows.
- Consent. If anyone other than you appears in the photos, get their agreement first.
Which kind of AI tool fits this job
Match the tool to the subject you want the model to learn.
Your face or personal photos. Photo AI turns a set of your selfies into a personal AI model, then produces photos of you in settings, outfits and poses you describe, with themed packs so you can skip writing prompts. MyPicNow trains a model of your face and generates portraits in more than 60 styles for LinkedIn, resumes and social media.
Products and brand imagery. Astria fine-tunes custom models from a person's or product's photos and exposes the capability through a REST API, with templates for fashion, beauty and jewelry. Flair AI trains custom product, human or aesthetic models from 6 to 10 images and places products into scenes with a drag-and-drop editor.
A style, character or artwork. Leonardo.Ai lets subscribers train personal AI models on their own references. Krea lists LoRA training on your own images to capture a style or subject. Playform lets artists train a model from a set of 30 or more images, and states that artists own the copyrights to what they create.
Step 1: Decide what the model should learn
A model trained on one clear subject beats one trained on a mixed folder.
- Choose one subject: you, one product, or one style.
- Write down where you will use the results, for example a profile photo, a catalog or a series of illustrations.
- Decide whether you need commercial rights, and check that when you choose a tool.
Check it worked by completing one sentence: "I want a model that can generate ___ in ___ settings."
Step 2: Collect and prepare your training photos
The quality of the photos sets the ceiling for the model.
- Gather the number of images the tool asks for. Flair AI lists 6 to 10 images, and Playform lists a set of 30 or more.
- Vary pose, expression, outfit, background and lighting for people, and angles for products. Avoid near-duplicates from a single shoot.
- Use sharp images. Astria recommends training images that are sharp and at least 512x512 pixels.
- Convert unsupported formats. MyPicNow accepts JPG, PNG, WebP and AVIF but not HEIC, so convert HEIC files to JPG first.
- Remove group shots, sunglasses that hide the face, heavy filters and anything explicit.
Check it worked by scrolling through the folder: every image should clearly show the subject and look different from the others.
Step 3: Choose a tool and plan
Pick the tool by subject, budget and how much control you want.
- Compare the training allowance. Leonardo.Ai lists personal model training of 10 to 50 per month by plan. Krea limits LoRA training to 50 images until the Max plan.
- Check what training costs. Playform lists training at 6 credits per hour, and Photo AI meters each action in credits.
- Look at the data terms. Astria deletes models, training photos and generated images after 30 days unless storage is extended.
- Use the comparison table below to see pricing and stated limits together.
Check it worked by confirming that your photo count and your commercial needs fit the plan you chose.
Step 4: Upload the photos and start training
Training turns your folder into a model you can prompt.
- Create the model or tune inside the tool and name it clearly.
- Upload the photos and, where asked, say what the subject is. Photo AI works from your selfies, MyPicNow asks for a mix of close-ups and full-body shots, and Astria works from a person's or product's photos.
- Confirm any consent or terms prompts, then start training.
- Wait for it to finish. MyPicNow says training takes roughly 30 minutes on average, while Photo AI says lower plans can take 30 minutes to 2 hours, so allow time.
Check it worked by seeing the model appear as ready in your account.
Step 5: Test with simple prompts
Start simple, so you can see whether the model learned the subject.
- Generate a few images with a short prompt that places the subject in a plain setting.
- Look at likeness, details such as hands or logos, and artifacts. MyPicNow notes that some images can contain minor artifacts such as distorted features.
- Try a themed pack if the tool offers one. Photo AI has packs such as dating profiles and social media.
- Add detail to prompts gradually: setting, outfit, lighting, style.
Check it worked by showing the results to someone who knows the subject and asking whether it looks right.
Step 6: Refine, upscale and save your results
Get the final images into shape and keep them safe.
- Pick the strongest results and regenerate close variations. Photo AI offers upscaling, relighting and uncrop tools, and Flair AI has an image editor with inpainting, object removal and upscaling.
- If the model is weak, retrain with better or more varied photos rather than writing longer prompts.
- Download the files you want to keep, since some services delete content after a fixed period.
- Check the commercial terms before you publish or sell anything.
Check it worked by opening a downloaded image at full size and confirming it is clean.
Tools for this job compared
| Tool | Good fit for | Pricing | Stated limits |
|---|---|---|---|
| Photo AI | Personal models from selfies, on the web | Paid, from $19/mo (checked 24 Oct 2026) | Credits per action and monthly credits expire; lower plans can take 30 minutes to 2 hours to train |
| MyPicNow | Headshots and social photos in 60+ styles | Paid, from $19.99/mo (checked 24 Oct 2026) | HEIC not supported; minor artifacts possible; refunds restricted once a subscription starts |
| Astria | Product and campaign imagery with fine-tuned models and an API | Paid | No public price list; content deleted after 30 days unless storage is extended |
| Flair AI | Custom product models and on-model fashion shots | Freemium, paid from $8/mo (checked 24 Oct 2026) | Mobile support limited; negative prompts not supported; unused credits do not roll over |
| Leonardo.Ai | Personal model training plus image and video generation | Freemium, paid from $12/mo (checked 24 Oct 2026) | Free plan allows 150 fast tokens per day and only the Auto model; free creations are public |
| Playform | Artists training on their own artwork | Freemium, paid from $10/mo (checked 24 Oct 2026) | Free generations are watermarked; training costs 6 credits per hour; credits cost $1 each |
| Krea | Style LoRAs with realtime canvas and upscaling | Freemium, paid from $9/mo (checked 24 Oct 2026) | Free plan capped at 100 compute units a day; LoRA training limited to 50 images until Max |
Prices and plan limits change, so confirm the current terms on each vendor's site before you subscribe.
Common mistakes to avoid
- Training on near-identical photos. Photos from one shoot teach the model one pose and one background. Fix it by varying lighting, expression, outfit and setting.
- Using blurry or small images. Low-quality inputs produce weak likenesses. Fix it by using sharp originals, and check the tool's size guidance, such as Astria's 512x512 minimum.
- Skipping consent. Training on a friend, colleague or celebrity without permission raises privacy and legal problems. Fix it by using only your own photos or those you have written permission to use.
- Ignoring how long files are kept. Some services delete models and images after a set period, and Photo AI's monthly credits expire. Fix it by downloading what you need and checking retention.
- Forgetting commercial terms. A free or low plan may not allow commercial use. Fix it by reading the licensing for your plan before you sell or publish the images.
How we put this together
This is an evidence-based guide built from each tool's listing in our directory, which is compiled from the vendor's own site and pricing page. It is not a review based on using the tools. We don't rank tools, and nobody paid to be included. Some outbound links may be affiliate links; see our disclosure.
Frequently asked questions
Is there a free way to train an AI image model on my own photos?
Partly. Playform and Krea have free tiers, though Playform's free generations carry a watermark and Krea's Free plan is capped at 100 compute units a day. Whether free plans include training varies, so read each plan before you upload.
How many photos do I need to train an AI image model?
It depends on the tool. Flair AI lists 6 to 10 images, and Playform lists 30 or more. Varied, sharp photos matter more than raw numbers.
How long does training take?
It depends on the service and plan. MyPicNow says roughly 30 minutes on average, and Photo AI says lower plans can take 30 minutes to 2 hours. Expect to wait before you can generate.
Can I keep my photos private?
Check each vendor's terms. MyPicNow says uploaded images are deleted from its servers within 30 days of removal, and Astria deletes models, training photos and generated images after 30 days unless storage is extended. Leonardo.Ai's free plan creations are public.
Can I train a model on someone else's face?
Only with their clear consent. Using a person's likeness without permission can cause legal and ethical problems, so keep to your own photos or get written agreement.
Compare more options in our Image generation category. To check whether an image was made by AI, see how to tell if an image is AI-generated.
