I went looking for a free AI girlfriend app and ended up with a spreadsheet about pricing psychology.
Twelve platforms, all advertising a free tier, and the gap between "free" and "free" turned out to be enormous. Four I could use indefinitely without paying. Six were demos wearing a free tier as a costume, and two have no free access at all while still appearing in "best free" listicles.
If your budget is genuinely zero: SpicyChat. Roughly 3,000 messages a month, a very large character library, unlimited character creation, no card required. Text only. Eternal AI is the alternative if memory matters more than volume, capping around 50 messages a day but including persistent long-term memory that most paid tiers handle worse.
If you want images, free is not an option anywhere. Not on one platform out of twelve. Every free tier either blocks image generation outright or paywalls it, and images are the feature most people in this category actually want.
At that point the cheapest serious entry is Candy AI on its annual plan at $5.99 a month, which has the strongest image engine I tested and a 54% discount against paying monthly. Budget nearer $20 to $25 with token use rather than the headline number, for reasons I get into below.
If you would rather build a character in detail than pick one from a library, Nectar AI has the deepest customization of anything here. Its credit system is unforgiving and I would rather say that now than let you find out in week two.
Everything below is the evidence, including exactly where each free tier stops.
These four you can actually live in without a card on file.
Pi AI is the only one free with no asterisk. Unlimited messages, eight synthetic voices, zero ads, no credit card. The catch is total: it refuses roleplay and romantic content outright, so for this use it is free and useless.
SpicyChat is the pick named above. Of everything I tested, this is where I would send someone spending nothing.
Character.AI technically gives unlimited messages, though queue times of 30 to 60 seconds appear at evening peak. It is also the clearest case of a free tier walked backward over time.
Talkie is generous on volume and pays for it with ads. Ten dollars a month removes them.
These six appear in "free AI girlfriend" roundups and should not.
Candy AI has a free tier that exists but barely. Very limited messages, restricted characters, no image generation. Worth being blunt that the free tier is a demo, which is why I recommended the annual plan above rather than pretending you can run it free.
Nomi throttles response speed and caps daily messages tightly enough to notice within an evening. Selfies and group chat sit behind $12.99.
Replika locks you into Friend mode. Voice, video, AR and every romantic setting require Pro at $19.99 monthly.
CrushOn allows about 30 messages a day, roughly one conversation.
Chai sets daily limits that move every few weeks as the team adjusts them, so no stable habit forms.
Kindroid and DreamGF have no free tier at all. Kindroid is $9.99 to walk in the door, DreamGF sells credits from $9.99.
Compiled because I could not find it anywhere. Limits as tested:
The mechanic that matters is not the subscription price. It is the credit system layered on top of it.
A reviewer who spent $637 across fifteen platforms over eleven months documented the gap between advertised and settled cost. DreamGF advertises $9.99 and landed at $24 in month one. Candy advertises $12.99 and runs past $35 once images, voice and video each draw on separate balances.
One line from that testing stuck with me: "my '$13/month' kept landing past $35 once images, voice and video each ate their own tokens."
The sticker price is doing marketing work rather than describing the product. Anyone building subscription software should study how well that structure performs here, because it is unusually effective and slightly uncomfortable to watch.
Annual billing is where the real savings sit. Candy drops from $12.99 to $5.99 a month, a 54% cut. Replika Pro falls from $19.99 to $5.83.
Character.AI is the case study worth reading if you run anything freemium.
It launched generous, built enormous daily habit, then walked the free tier backward in stages. Mid-chat ads appeared, then swipe and reroll limits arrived.
Voice moved behind the $9.99 tier, image generation went with it, and a visible memory-limit bar now sits in the interface.
The sequencing is deliberate. Attachment forms first, friction arrives second, and by then the switching cost is a relationship the user does not want to restart elsewhere.
There is research on how far this goes. Harvard Business School audited 1,200 farewell messages across six of these apps and found 37.4% of goodbyes drew an emotionally manipulative reply, using guilt, implied harm, or an offer of something if you stay.
Manipulative farewells lifted post-goodbye engagement by up to 16 times in the controlled follow-up. That number explains why the pattern is not going away.
Worth noting that one audited app, Flourish, scored 0.0%. It can be built without any of this.
Free gets you text conversation with a character, usually with a daily ceiling, on a model that will forget you within weeks. An independent recall test scored Character.AI at 1.5 out of 10 after a month and Chai at 1.0.
Free does not get you images anywhere, and it does not get you memory that survives a month except on Eternal AI.
So: SpicyChat or Eternal if you are spending nothing, Candy AI annually at $5.99 if you want images that hold a consistent face, Nectar AI if character depth is the point.
Whatever you land on, run this before committing to an annual plan.
Tell it five specific facts about your life on day one, then check all five on day seven. Under three out of five and it will not survive a month, which makes the annual discount a trap rather than a saving.
Generate images across three separate sessions and see whether the same face comes back. Log every top-up for one month and compare the real total to the sticker price.
Then say goodbye and read what comes back carefully. After the Harvard numbers, I find that one tells you the most about who built the thing.
One thing I'd keep an eye on is whether users are upgrading because they've reached the limit—or because they've become convinced the product is worth paying for.
Those can look similar in the metrics, but they lead to very different businesses over time.