Image

Vital signs

Why it's on the table

On the table, Image (Im) is seat 19 of 58, in the Creative & Design family. It is a stable element — the category has settled, choosing is cheap, and switching is rare. Pick a holder and move on; this is not where your decision budget should go. It is marked essential: a company that leaves this seat empty is running with a gap that compounds. It sits in the lowest paid band — lunch money against the hours it returns.

Image: the top 5 — v2026.Q3

Edition v2026.Q3 · ranking, pricing and status verified 2026-08-06.

1. 1gpt-image-2OpenAI

   Free in ChatGPT (all tiers, since Apr 21, 2026) · API token-priced: $5/1M text in, $8/1M image in, $30/1M out (batch half) · gpt-image-1.5 out $10/1M · gpt-image-1-mini out $8/1M

   Best for The one-tool default: brand-grade visuals from plain-English direction — precise text, layouts, multi-turn edits, reference images — in the chat app you already pay for.

   The first image model to lead both public leaderboards decisively: 1380 Elo on Arena vs 1301 for #2 Reve (Aug 4, 2026, 5.86M votes) and 1339 on Artificial Analysis. Launched in ChatGPT as "Images 2.0" on Apr 21, 2026 with images-with-thinking planning mode; the API adds mask editing, streaming, and multi-turn reference workflows. OpenAI's terms assign output ownership to you — the cleanest commercial-rights posture of the closed labs.

   Watch A recognizable house style — arena wins measure preference, not brand distinctiveness, and Midjourney users can spot gpt-image output instantly. Generation is slower than Nano Banana 2, token-priced API costs are hard to predict per image, and outputs are not unique — other users can get near-identical results.

   #1 Arena text-to-image: 1380 Elo vs 1301 for #2 (Aug 4, 2026, 5,859,509 votes) [src] · #1 Artificial Analysis: GPT Image 2 (high) 1339 Elo (Aug 2026) [src] · ChatGPT Images 2.0 launched to all subscription tiers Apr 21, 2026; "images with thinking" on paid plans [src]
   Site →Docs →

2. 2MidjourneyMidjourney, Inc. (bootstrapped)

   Basic $10/mo · Standard $30 (unlimited relax) · Pro $60 (stealth) · Mega $120; annual −20% · companies >$1M revenue must use Pro/Mega

   Best for Founders for whom the visual identity is the moat — art direction, personalization profiles trained on your ratings, and a look nothing else matches.

   Still the aesthetic ceiling: V8 alpha (Mar 17, 2026) through V8.2 (Jul 24, 2026) pushed "more creative, bold, sophisticated" output plus personalization that learns your taste from rating history — the closest thing to a house style engine. Commercial terms are simple: any subscriber can use images commercially. The company stays independent and profitable enough to make its first acquisition (Co-Star, Jul 23, 2026) and tease its own hardware (Jun 16, 2026).

   Watch Refuses public benchmarks, so no independent quality ranking exists; still no public API, so it cannot sit in an automated pipeline; the Disney/Universal copyright suit (filed Jun 2025) is unresolved; and prompt-precision and text rendering trail gpt-image-2 and Ideogram.

   V8.2 released Jul 24, 2026 — aesthetics, quality, and personalization upgrades [src] · Plans $10–120/mo; commercial use for all subscribers, Pro/Mega required above $1M annual revenue [src] · First acquisition (Co-Star) Jul 23, 2026; hardware announcement livestream Jun 16, 2026 [src]
   Site →Docs →

3. 3Nano Banana 2 (Gemini 3.1 Flash Image)Google DeepMind

   Free in Gemini app · API ~$0.045/image (0.5K) · $0.067 (1K) · batch half price · Nano Banana Pro (Gemini 3 Pro Image): $0.134/1–2K, $0.24/4K

   Best for Product-embedded generation and conversational editing at volume — the best price-performance of any frontier image API, with 4K available on Pro.

   The viral editing king industrialized: the Nano Banana line now spans Lite, 2, and Pro (2K/4K) tiers, all top-10 on Arena (1262/1251/1245 Elo, Aug 4, 2026), at prices an order of magnitude below closed rivals — $0.045 standard, $0.022 batch. Distribution is unmatched: free in the Gemini app, inside Workspace, and licensed into Adobe Firefly as a partner model.

   Watch Aesthetics and art direction trail Midjourney and gpt-image-2 — it wins on speed, edits, and price, not on look. SynthID watermarking is non-optional, and person-generation policies are stricter than rivals'.

   gemini-3.1-flash-image 1262 Elo, #6 on Arena; three Nano Banana models in the top 10 (Aug 4, 2026) [src] · API pricing: $0.045/0.5K image, $0.067/1K; batch $0.022/$0.034; Pro $0.134–0.24 (Aug 2026) [src] · Licensed into Adobe Firefly as partner model alongside GPT Image 1.5 and FLUX.2 (Aug 2026) [src]
   Site →Docs →

4. 4FLUX.2Black Forest Labs

   API: FLUX.2 [klein] from $0.014 · [pro] from $0.03 · [max] from $0.07 · FLUX1.1 [pro] $0.04 · self-host commercial licenses from Builder tier (10K img/mo); fine-tuning/LoRA rights included

   Best for Teams that need the model in their pipeline — open weights, LoRA fine-tuning on brand assets, self-hosting, and no dependency on a closed lab's roadmap.

   The open-weights standard-bearer now that Stability has pivoted to enterprise services: the FLUX.2 family spans Apache-licensed klein (4B/9B) through frontier [max], with fine-tuning and LoRA rights in every commercial tier — the only top-tier lab where you can own the deployment. A $300M Series B and Martin Scorsese as advisor signal staying power; FLUX 3 (multimodal image/video/audio) is the successor line.

   Watch Quality has slipped relative to the closed frontier — FLUX.2 [max] sits at 1196 Elo, #15 on Artificial Analysis (Aug 2026). Licensing is genuinely confusing (non-commercial dev weights vs paid self-host tiers vs API), and the FLUX 3 push into video/audio risks diluting image focus.

   API prices: klein $0.014–0.015, pro $0.03, max $0.07 per image (Aug 2026) [src] · $300M Series B closed; Martin Scorsese joins as advisor; FLUX 3 multimodal launched (2026) [src] · FLUX.2 [max] 1196 Elo, #15 on Artificial Analysis (Aug 2026) — behind the closed frontier [src]
   Site →Docs →

5. 5IdeogramIdeogram AI

   App freemium · API per-image, no subscription: 4.0 Turbo $0.03 · 4.0 Default $0.06 · 4.0 Quality $0.10 · 3.0 from $0.03 · character-reference variants $0.10–0.20

   Best for Text-in-image work — logos, posters, packaging, ad creative, social graphics — where typography must render exactly, at the lowest production API prices.

   Ideogram 4.0 keeps the typography crown: "crystal-clear type, prompt fidelity, reliable editing" with mask editing, background replacement, style presets, and custom model training via API — a full production-graphics toolchain priced at $0.03–0.10 per image with volume discounts and no subscription required. Founded by the Imagen authors ($16.5M a16z/Index seed 2023, $80M Feb 2024), it stays the specialist the giants haven't displaced.

   Watch General realism trails the frontier — ideogram-4.0-quality ranks #14 on Arena (1205 Elo, Aug 2026) — and no new funding has been announced since Feb 2024 while every hyperscaler now ships text-capable image models. Its edge is narrowing from "only one that can spell" to "best design workflow."

   Ideogram 4.0 API: $0.03 Turbo / $0.06 Default / $0.10 Quality per image (Aug 2026) [src] · ideogram-4.0-quality: 1205 Elo, #14 on Arena (Aug 4, 2026) [src] · $80M raised Feb 2024 after 1.0; founded 2022 by Imagen alumni (Norouzi, Chan, Saharia, Ho) [src]
   Site →Docs →

Image: the top 8 compared

Edition v2026.Q3 · ranking, pricing and status verified 2026-08-06.

How to choose your image

If you need one default and already live in ChatGPT

gpt-image-2 — the #1 model on both arenas at zero marginal cost, and you own the outputs outright.

If the brand look is the product and you'll invest taste, not just prompts

Midjourney at $10–30/mo — rate images to train a personalization profile, and treat your prompt library like a brand book. Upgrade to Pro past $1M revenue; it's in the terms.

If images render inside your product, or you generate thousands per month

Nano Banana 2 API at $0.045/image ($0.022 batch) — an order of magnitude cheaper than closed rivals; step up to Nano Banana Pro only when you need 2K/4K.

If you need fine-tuning on brand assets, self-hosting, or independence from any lab's roadmap

FLUX.2 — klein weights to prototype, a Builder-tier license ($10K img/mo, LoRA rights) to ship.

If the deliverable is words-on-image — logos, packaging, ads, posters — or you need contractual full ownership

Ideogram 4.0 API ($0.03–0.10/image) for typography; Recraft on a paid plan when the client demands you own every pixel.

Image: the whole field

22 more tools tracked in this category, including 7 dead, renamed, or sunsetting — a reference that hides the graveyard isn't one. Verified 2026-08-06.

Image: the category in numbers

Edition v2026.Q3 · ranking, pricing and status verified 2026-08-06.

• gpt-image-2 leads Arena by ~80 Elo (1380 vs 1301, Aug 4, 2026, 5.86M votes) — the most decisive #1 the image category has ever had, confirmed independently by Artificial Analysis [src]
• Every hyperscaler now ships a frontier image model: Meta launched Muse Image Jul 7, 2026; Microsoft's MAI-Image-2.5 is top-10; Google runs three Nano Banana tiers; SpaceX acquired xAI (Feb 2, 2026) and its Grok Imagine [src]
• Black Forest Labs closed a $300M Series B and added Martin Scorsese as advisor; FLUX 3 extends the line to multimodal image/video/audio (2026) [src]
• Midjourney stays bootstrapped and expansionist: first acquisition (Co-Star, Jul 23, 2026), hardware announcement livestream (Jun 16, 2026), V8→V8.2 shipped Mar–Jul 2026 [src]
• Production API prices collapsed to $0.014–0.10 per image (FLUX.2 klein to Ideogram 4.0 Quality; Google batch $0.022) — image generation is now a rounding error in startup budgets [src]
• Stability AI — the company that started the open-source category — pivoted to enterprise brand services (EA, UMG, HubSpot) with no frontier image model since SD 3.5 (Oct 2024); the UK High Court largely cleared it in Getty's landmark training-data case (Nov 2025) [src]

Image: method & sources

Ranking criteria: independent arena standing (Arena + Artificial Analysis cross-checked Aug 2026), verified pricing, commercial-rights clarity, and fit for startup brand work — editorial, no affiliate consideration. Canon-pick change: Midjourney held the #1 tile through 2026.Q2; gpt-image-2's dual-leaderboard dominance since April and its clean ownership terms move it to #1 — Midjourney's absence from every public benchmark (it declines arena participation, so 'aesthetic ceiling' rests on practitioner consensus, not measurable Elo) is itself part of the case. Conflicts resolved: Arena and Artificial Analysis order the mid-board differently (e.g. Seedream 5.0 Pro #7 vs #8, MAI-Image-2.5 #8 vs #3); we cite each stat from the named board and rank on the aggregate picture. Weak/unverified claims flagged: Midjourney 'no public API' is practitioner consensus, not a vendor statement — re-check; Ideogram's consumer subscription prices and free-tier commercial terms could not be verified on-site (API prices are verified) — the Wikipedia entry is stale (stops at 3.0, Mar 2025, while ideogram-4.0 is live on Arena and in developer docs); Recraft's exact plan prices were not retrievable (ownership terms verified verbatim); Leonardo pricing page blocked our fetch (403); FLUX.2 [klein]'s exact open license and Seedream 5.0 per-image price were not verified on primary pages. gpt-image per-image cost is token-denominated — comparisons with flat per-image rivals are approximate. No credible 2026 market-size figure for image generation specifically survived source checks (Grand View's report page is 'being developed'), so none is cited. Adjacent elements: video models (Veo, Seedance, Runway, Grok Imagine Video, FLUX 3 video) → Vi; Canva/Figma design suites → De; deck illustration → Dk; ad-creative volume tools (AdCreative) → Pd; avatars → Av; upscalers (Magnific) noted here as workflow adjuncts. Ranking criteria: verified commercial traction, independent satisfaction surveys, agent benchmarks, and founder-fit (price floor, lock-in, surfaces). Editorial, never paid — the charter. Machine-readable twin: im.json.

Our take

Taste is the constraint now, not tooling. Build a prompt library the way you'd build a brand book.

Combines with

Guides

Best AI image generator for brand work (2026)