API pricing cheat-sheet (Q1 2026)
Headline per-image prices have compressed 5–8× in 18 months. The differentiator is now quality, typography, and aspect-ratio flexibility — not raw cost. DALL-E 3 and GPT-Image-1 win on prompt adherence; Midjourney v6 wins on aesthetic quality and brand vibe; Ideogram wins on legible text-in-image; Imagen 4 wins on photorealism and humans; Stable Diffusion XL wins on cost when self-hosted. Sora and Veo handle short video frames at $0.50–$2.00/clip.
- • DALL-E 3 HD 1024: $0.080/image (OpenAI API)
- • GPT-Image-1 medium: $0.04/image (OpenAI API)
- • Ideogram 2.0 Turbo: $0.04/image
- • Imagen 4 standard: $0.04/image (Vertex AI)
- • Midjourney Pro subscription: $60/mo for ~5k images = $0.012/image
- • SDXL self-hosted on L40S: ~$0.0008/image at full GPU utilization
Hidden costs the headline pricing misses
Three line items eat budgets quietly. (1) Regeneration tax — a 60% reject rate means you pay 2.5× the headline cost per kept image. (2) Human review — even at $40/hr and 2 minutes per image, that's $1.33/kept image, often 10–50× the API cost itself. (3) Failed runs from API errors, rate limits, and timeouts add a 3–8% surcharge in production environments.
When to self-host SDXL or Flux
Self-hosting on a single rented L40S or A10G ($0.45–0.85/hr) makes sense above roughly 1,500 images/day or when latency matters (gaming, real-time UX). The break-even vs DALL-E 3 HD is ~150 images/day at 24/7 GPU rental. Add 0.5 FTE of MLOps overhead and your break-even is realistically 3,000+ images/day. Below that, just pay the API and spend the engineer time on product.
Reducing the reject rate is the highest-ROI lever
A 20-point drop in reject rate (from 60% to 40%) cuts API spend ~33% with no model change. Three techniques deliver this: (a) reference image conditioning to lock style and composition, (b) negative prompts to remove your most common failure modes, (c) seed + parameter logging so successful generations can be replicated at scale. Worth investing 1–2 weeks of prompt engineering before scaling volume.
Related guides
Long-form playbooks on the same topic, written by the RevenueLab editorial team.
FAQ
Is Midjourney cheaper than DALL-E?
Per-image at full utilization, yes — Midjourney Pro at $60/mo delivers ~5,000 fast-mode images ($0.012/img) vs DALL-E 3 HD at $0.08/img. But MJ has no API and requires Discord or unofficial wrappers; DALL-E is a clean OpenAI API call. Pick on workflow fit, not price alone.
How does Sora pricing work?
Sora pricing is per-second of output and resolution-dependent. As of early 2026, OpenAI is in early access tiers; expect $0.50–$2.00 per 5-second clip. Use our AI video generation cost calculator for video-specific math.
Do I pay for failed or filtered images?
Most providers charge for any generation that completes successfully, even if the content filter blocks it client-side. Failed-on-server generations (timeout, model error) are typically not billed. Build a 3–5% buffer for filter rejection in safety-sensitive use cases.
Can I commercially use AI-generated images?
Depends on provider terms and jurisdiction. OpenAI, Midjourney, Ideogram, and Stability grant commercial rights to paid-tier output. The U.S. Copyright Office does not register purely AI-generated work without human authorship modifications. Consult counsel for any campaign at scale.
What about training data and IP risk?
Adobe Firefly and Getty Generative AI offer indemnity. Most others do not. If you're producing client work or large ad spend, the legal premium for an indemnified model often pays for itself in one cease-and-desist saved.
How do I cut cost on product photography at scale?
Generate background and lighting variants of a single real product shot via image-to-image conditioning at $0.04/img instead of re-shooting. Typical savings: 80–95% vs a studio shoot, with comparable quality for ecommerce thumbnails.
Is the cheapest model always best?
No. SDXL at $0.0008/img sounds great until you realize it needs 3–5× more regens than DALL-E 3 for the same accepted image, plus self-host overhead. Always price by 'cost per kept image,' not headline rate.
What about local generation on a Mac M3 Max?
Viable for <100 images/day creative workflows. Free after hardware, but slower (15–30s/img) and limited to SDXL/Flux quality. Not viable for production volume.
How this calculator is built
Independently maintained
Written by Sam Doshi and the RevenueLab editorial team. We don't sell the data feeds this tool is built on.
Sourced from primary data
Benchmarks come from public AdSense / Stripe / IRS disclosures and reader-submitted data — never third-party "$X per view" claims. Full methodology.
Last reviewed
June 2026. We re-check every figure on the platform on a rolling quarterly cycle.
Editorial standards
See our editorial policy and disclaimer. Results are estimates, not advice.