How to generate AI images free, every day

Most "free" AI image generators hand you a block of trial credits once and then stop. This one refills. The free plan on OpenModels carries a wallet of green credits that tops itself back up every hour toward a daily cap, and that wallet pays for the models running on our own GPUs. Here is exactly what it covers, and exactly where it runs out.

What "free" means here

There are two wallets on an OpenModels account and they are not interchangeable. Green credits (⚡) refill on a clock and can only pay for models that run on hardware we own. Gold credits (✦) are bought with money and pay for everything, including the hosted models from Google, ByteDance, OpenAI and the rest. Signing up grants zero gold.

That split is the whole reason a free tier exists at all. A generation on our own GPU costs us electricity on a machine already paid for, so it can be generous. A generation on a hosted API costs real dollars per call, so it cannot.

Which plan: A free account's green wallet caps at ⚡20 a day and refills continuously toward that cap — about one credit every 72 minutes rather than a lump at midnight. Unused balance does not stack; it fills to the cap and stops.

What each plan's green wallet is capped at

Plan	Green credits a day	Parallel jobs
Free	20	2
Starter	100	3
Creator	200	5
Pro	400	10
Ultra	2,000	100

Which models the free plan reaches

The own-GPU image lane is what green credits buy. Every model below runs on our hardware and starts at ⚡1 an image.

• Z-Image Turbo — the fast general-purpose model. This is the default, and the one a signed-out visitor's single free picture runs on.
• Qwen Image Edit (Local) — edits a photo you attach, from a written instruction.
• Krea 2 Identity — restages a person or product you supply while keeping them recognisable.
• Krea 2 Turbo — a distilled, faster Krea. It ignores negative prompts, because the turbo checkpoint samples at CFG 1 and a negative would be encoded and then discarded.

What it costs: Prices on our own hardware start at ⚡1 an image. They are set at runtime and can carry a demand multiplier when the render queue is deep, and if the local lane is unavailable the same model is served over a paid API and billed in gold instead. Treat ⚡1 as a floor, not a fixed price.

GPT Image 2, which costs nothing at all

One hosted model is genuinely free on every plan: GPT Image 2 runs at 0 credits and accepts up to six reference images. The daily allowance rises with the plan — three runs a day on Free, ten on Starter, twenty-five on Creator, seventy-five on Pro and two hundred on Ultra. It rides a shared subscription rather than a metered API key, which is also its limit — when that subscription's daily allowance is exhausted the model pauses for everyone until it resets. It is a real option, not a way to avoid spending credits.

How to generate your first free image

1. Open the composer. Go to /image. The prompt bar sits at the bottom; the model picker is the chip above it.
2. Pick an own-GPU model. Z-Image Turbo is selected by default on a free account. Models your plan cannot run are shown with the plan that opens them, rather than hidden.
3. Write the prompt. Describe the subject, then the framing, then the light. "A ceramic mug on a windowsill, shot from slightly above, hard morning light" beats "beautiful mug photo" every time.
4. Set the batch. A free account runs two jobs at once, so asking for four copies gets you two. Start with one until the prompt is right.
5. Press Generate. The cost is quoted on the button before you press it. Results land in the gallery and in your assets library.

Try it with a starter prompt — Opens the composer with this prompt loaded. Nothing runs until you press Generate.

How to make ⚡20 go further

The wallet refills at a fixed rate, so the only real variable is how many credits each attempt costs you. Four habits account for most of the difference between twenty useful pictures a day and five.

• Iterate at batch 1. A batch of two costs twice as much and tells you almost nothing extra while the prompt is still wrong. Get the composition right on single images, then raise the batch once you are only changing details.
• Fix the prompt before you change the model. Switching models to escape a bad prompt spends credits proving that the prompt was the problem. Z-Image Turbo is fast enough that a second attempt costs you a minute.
• Use Qwen Image Edit rather than regenerating. If a picture is nearly right, editing it from an instruction keeps everything you already liked. Starting again re-rolls the parts that were working.
• Spend GPT Image 2's daily runs first. They cost no credits at all, so using them before touching the green wallet is free reach. Three a day on Free, and the allowance rises with every plan.

One thing that does not help: waiting for a quiet moment to get a cheaper price. The demand multiplier applies to what a render costs while the queue is deep, and it is visible on the Generate button before you press it — but the base rate is set by us, not by the clock, so there is no cheap hour to wait for.

What these models respond to

The own-GPU models are smaller than the hosted frontier ones, and that changes what a good prompt looks like. They follow concrete, physical description well and abstract art-direction badly. "Shot on 35mm, shallow depth of field, overcast light from the left" lands; "evocative, award-winning, masterpiece" does almost nothing except spend a credit.

Order matters more than length. Put the subject first, then what it is doing, then where it is, then the light, then the camera. If a detail is essential, say it once and plainly rather than three times in different words — repetition does not weight it, it just crowds out the rest of the description.

Where the free plan stops

Being straight about this is more useful than a longer feature list. The free plan stops in five specific places.

Limit	On the free plan	What lifts it
Daily credits	⚡20, refilling	Any paid plan raises the cap
Hosted models	Not reachable with green credits	Buying gold credits, or a plan that grants them
Watermark	Always on	Creator and up can switch it off
Parallel jobs	2 at a time	Starter 3, Creator 5, Pro 10
Storage	50 MB	Every paid tier raises it

Watch out: Green credits are never for sale. Buying credits buys gold, which pays for hosted models and queue priority — it does not raise your daily green allowance. Only the plan does that.

What if it doesn't work?

• "Not enough credits" — the green wallet refills hourly rather than all at once, so waiting is often the whole fix. The wallet chip in the header shows the current balance.
• The model you picked is greyed out — it is a hosted model, which green credits cannot pay for. The chip names the plan or the wallet that opens it.
• The job is queued behind others — our GPUs run one job at a time per slot and the free lane is the slowest. The queue board shows the real wait.
• GPT Image 2 is unavailable — its shared allowance is spent for the day. The own-GPU models are unaffected.

Is it really free, or a trial?

It is a recurring allowance, not a trial. The green wallet refills toward its daily cap every hour, for as long as the account exists. There is no card required and nothing expires into a paywall.

How many images is ⚡20 a day?

Roughly twenty on the own-GPU lane, where images start at ⚡1. Prices are set at runtime and can rise when the render queue is deep, so treat it as about twenty rather than exactly twenty.

Can I use the free images commercially?

Yes, subject to the Terms. Output on the free and Starter plans carries an OpenModels watermark, which Creator and higher plans can switch off.

Can I reach Nano Banana Pro or Seedream for free?

No. Those are hosted models billed per call by their provider, and green credits cannot pay for them. They open from the Starter plan, paid in gold credits.

Do I need an account?

For the daily allowance, yes. Without one you can still generate a single real picture on Z-Image Turbo before signing up, and it is kept and moved to your account when you do.