Free tool
AI Image Generation Cost (2026)
What does it actually cost to make AI images? Compare flat subscriptions (Midjourney, Ideogram, Leonardo, Firefly), pay-per-image APIs (OpenAI gpt-image, Google Imagen), and free self-hosting (Stable Diffusion, FLUX) — with commercial-use rights. Verified June 2026.
- SubscriptionMidjourneyBasic$10/mo
~3.3 fast GPU hrs ≈ ~200 images; no relax mode
- Commercial: Yes (companies >$1M rev need Pro/Mega Stealth)
Best-in-class aesthetic quality
- SubscriptionMidjourneyStandard$30/mo
15 fast hrs ≈ ~900 images + UNLIMITED relax
- Commercial: Yes
Unlimited relax mode — best value tier
- SubscriptionMidjourneyPro$60/mo
30 fast hrs ≈ ~1,800 + unlimited relax
- Commercial: Yes
Adds Stealth (private generations)
- SubscriptionMidjourneyMega$120/mo
60 fast hrs ≈ ~3,600 + relax + Stealth
- Commercial: Yes
Highest fast-hour allotment
- SubscriptionIdeogramFree$0
~10 slow credits/week
- Commercial: Limited (paid grants rights)
Best in-image text/typography
- SubscriptionIdeogramBasic$8/mo
400 priority + ~100 slow credits/day
- Commercial: Yes
Cheapest paid entry, strong text
- SubscriptionIdeogramPlus$20/mo
1,000 priority + unlimited slow
- Commercial: Yes
Unlimited slow + priority queue
- SubscriptionIdeogramPro$60/mo
~3,000 priority + unlimited slow
- Commercial: Yes
High-volume priority credits
- SubscriptionLeonardo.AiFree$0
150 fast tokens/day (no rollover)
- Commercial: No (personal use only)
Generous free + game-asset/LoRA tools
- SubscriptionLeonardo.AiApprentice$12/mo
8,500 tokens/mo + 1 LoRA train
- Commercial: Yes
Consistent Character + model training
- SubscriptionLeonardo.AiArtisan$30/mo
25,000 tokens + 20 model slots
- Commercial: Yes
Mid-tier volume + more slots
- SubscriptionLeonardo.AiMaestro$60/mo
60,000 tokens + 50 LoRA slots
- Commercial: Yes
Highest token + LoRA allotment
- SubscriptionAdobe FireflyFree$0
25 generative credits/mo
- Commercial: Yes (commercially safe)
Licensed training data + indemnification
- SubscriptionAdobe FireflyStandard$9.99/mo
Unlimited standard image gens
- Commercial: Yes (safe + indemnified)
Unlimited standard at the lowest paid tier
- SubscriptionAdobe FireflyPro$29.99/mo
Unlimited standard + larger premium pool
- Commercial: Yes (safe + indemnified)
Premium/video credits + partner models
- Per-image APIOpenAIgpt-image-1 · Low (1024²)$0.011/image
Per image; batch ≈ half price
- Commercial: Yes (you own outputs)
Cheapest OpenAI; strong prompt-following
- Per-image APIOpenAIgpt-image-1 · Medium$0.042/image
Per image; batch ≈ half price
- Commercial: Yes
Balanced quality/cost for apps
- Per-image APIOpenAIgpt-image-1 · High$0.167/image
Per image; larger sizes cost more
- Commercial: Yes
Highest fidelity of gpt-image-1
- Per-image APIOpenAIgpt-image family · range$0.005–0.25/image
mini / 1 / 1.5 / 2 across quality+size
- Commercial: Yes
gpt-image-2 flagship; mini ~$0.005 undercuts all
- Per-image APIGoogle ImagenImagen 4 Fast$0.02/image
Pay-as-you-go on Google Cloud
- Commercial: Yes
Cheapest Imagen; speed-optimized
- Per-image APIGoogle ImagenImagen 4 Standard$0.04/image
Per image
- Commercial: Yes
Balanced default quality
- Per-image APIGoogle ImagenImagen 4 Ultra$0.06/image
Per image; free dev tier via AI Studio
- Commercial: Yes
Highest-fidelity Imagen tier
- Self-hostStable DiffusionSD 1.5 / SDXL (local)free
Unlimited local; ~8–16 GB VRAM
- Commercial: Yes (OpenRAIL-M, no rev cap)
Zero per-image cost + huge LoRA ecosystem
- Self-hostStable DiffusionSD 3 / 3.5 (local)free <$1M rev
Unlimited local generations
- Commercial: Free under $1M/yr org rev; else paid license
Newer arch; watch the $1M licensing gate
- Self-hostFLUX (Black Forest Labs)FLUX.1-schnell (local)free
Unlimited local; fast distilled model
- Commercial: Yes (Apache 2.0, unrestricted)
Open + commercially free + fast — great default
- Self-hostFLUX (Black Forest Labs)FLUX.1-dev (local)free (non-commercial)
Unlimited local; ~16 GB VRAM (8 with FP8)
- Commercial: Non-commercial; commercial self-host needs a BFL license
Top open-weight quality (mind the license)
Verified June 2026. Subscription image counts (esp. Midjourney) are ESTIMATES — Midjourney sells fast GPU time (~1 min/image), not a fixed count; Ideogram/Leonardo meter in credits/tokens where one image costs a variable amount. Per-image API and subscription prices aren't directly comparable without an assumed monthly volume. Self-host “free” = $0 software only; you pay for the GPU — and commercial rights differ by model (FLUX.1-dev is non-commercial without a license). Confirm current pricing with each vendor.
Want the verdict, not just the table?
The table shows every price; the guide adds the opinionated pick — which to choose for quality, for building into an app, or for free self-host.
Frequently asked
Subscription vs per-image API — which is cheaper for AI images?
It depends entirely on volume. Per-image APIs (OpenAI gpt-image from ~$0.005 with gpt-image-1-mini, Imagen 4 from $0.02) win for low or spiky usage because you pay nothing when idle. Flat subscriptions (Midjourney $10–$120, Leonardo $12–$60, Ideogram $8–$60) win for steady high-volume creative work — Midjourney Basic at $10 for ~200 images is ~5¢/image, and on its unlimited relax mode the effective per-image cost approaches zero, which no per-image API can match.
What's the absolute cheapest way to generate AI images?
Self-hosting an open-weight model (Stable Diffusion SD1.5/SDXL, or FLUX.1-schnell) is effectively free per image — you only pay for your own GPU/electricity or a rented cloud GPU. The cheapest hosted per-image rate is OpenAI's gpt-image-1-mini at roughly $0.005 and gpt-image-1 low quality at $0.011, followed by Google Imagen 4 Fast at $0.02.
How many images does a Midjourney plan actually give me?
Midjourney sells fast GPU TIME, not a fixed image count. A typical still image takes ~1 GPU-minute, so Basic's 3.3 fast hours ≈ ~200 images, Standard's 15 hours ≈ ~900, Pro's 30 ≈ ~1,800, Mega's 60 ≈ ~3,600. These are estimates that shift with model, upscales and settings — and Standard/Pro/Mega also include unlimited (slower) relax-mode generations, so the practical ceiling is far higher.
Which AI image services are safe for commercial use?
Paid Midjourney, OpenAI gpt-image, Google Imagen, paid Ideogram, paid Leonardo, and Adobe Firefly all grant commercial rights — Firefly goes furthest with commercially-safe training data and IP indemnification. For self-host: SD1.5/SDXL and FLUX.1-schnell (Apache 2.0) are commercial-free; SD3/3.5 is free commercially only under $1M/yr org revenue; FLUX.1-dev needs a paid Black Forest Labs license for commercial self-hosting. Free tiers (Leonardo, Ideogram) are generally personal-use only.
Why do Ideogram and Leonardo show 'credits' or 'tokens' instead of images?
Both meter usage in an abstract currency where one image costs a variable number of units depending on model and quality. On Ideogram, a top-quality image costs ~6 credits while a turbo image costs a fraction of one — so a 1,000-credit plan could yield anywhere from ~167 to thousands of images. Leonardo tokens work the same way, and Ideogram priority credits expire monthly with no rollover.