Every figure here carries the date it was read and the source it came from. How scoring works →
AI Image Generators, Measured on Output, Crowd, Price and Contract Terms
5 tools measured · Vouch Score data collected 17 June 2026 · arena re-read 5 September 2026 · highest composite first
Every image tool on this page is held to one measuring stick: public arena preference data for output quality, review-crowd aggregates for usability, verified pricing, and four clause positions read off each vendor's own terms: who indemnifies whom, what the paid tier lets you own, what the free tier permits, and the published refund position.
Top score: Recraft. It holds the highest Vouch Score composite on this page, 81/100, measured 17 June 2026, arena re-read 5 September 2026. The label is computed from the cards, not chosen by us, and it moves the day another tool measures higher. Read what each card measured before you take it as advice for your own work.
1Recrafttop score · highest composite81/100 · high
Vector output and brand assets
Vector-first generator aimed at brand assets and layout-safe design output.
Recraft is a vector-first generator aimed at brand assets: logos, icons, stickers,
book illustrations, game art. It is not a general image tool. Its raster model is strong
(arena ELO 1019, V4.1 Utility Pro, read 5 September 2026), and native SVG output is the real
differentiator. Our composite: 81.0/100, rebuilt 12 August 2026 from Recraft's own
terms. Paid from $12/mo; the free tier is non-commercial and §7.1 keeps its output.
Aesthetic-driven image generator with a devoted community and no free tier.
The Midjourney rating split is real, and both sides are right. Trustpilot's 1.6/5
judges Midjourney, Inc.: billing, refunds, support; G2's 4.4/5 judges the model's
output. The card composites 80.0/100, capability 71.6 from an arena ELO of 897. Since
18 August 2026 the consumer record sits beside that number rather than inside it:
Trustpilot has no recent history of asking for those 366 reviews, so they are reported
rather than averaged in. Terms read 12 August 2026.
Production-oriented image platform with model fine-tuning and asset pipelines.
Leonardo AI scores 75.5/100 (3.8/5), and it fell two points on 18 August 2026 when its
largest review record left the usability row. Its terms were read in full on 12 August: 38,524 characters, zero mentions of indemnity in either direction, and
§8.7 assigning ownership of free-tier output to Leonardo. The Apprentice plan is
still genuinely free, where Midjourney has none. One honest gap: hands still fail.
Arena read 5 September 2026.
Image generator known for reliable text-in-image rendering.
Ideogram is the AI image generator built around text-in-image: legible, correctly
spelled words, logos and multi-line layouts that rivals like Midjourney miss. Our
composite is 75.3/100 across all four rows, and capability rests on an arena ELO of 1015,
4.0-era, read 5 September 2026. The fourth row reads Ideogram's own terms rather than a star
rating. You own your output, and subscriptions are non-refundable.
Realtime-draft image generator built for fast visual iteration.
Krea AI is a realtime image generator: paint or type and the model refines the
canvas live, with a built-in upscaler and enhancer. Our composite is 74.8/100
(3.7/5), rebuilt 12 August 2026. All four rows carry a figure, and the fourth is the
one that fell: read in full, Krea's terms never settle who owns the output you pay
to make. Capability 82.9 comes from an arena ELO of 1015, read 5 September 2026. Billing
is the recurring complaint.
Outbound links may be affiliate links and can earn us a commission: they never touch a score, and the order on this page is the composite order, computed at build time from the cards themselves.
The score cards in full
What each card measured and what it did not: every number dated, sourced and reproducible.
Free tier + paid from $9/mo, or $5/mo billed annually (Basic)
Read the dimensions, not the composite
The composite hides trade-offs, so read the dimensions underneath it. Midjourney's output draws affection in its product reviews while a consumer-trust record of 366 reviews at 1.6 out of 5 sits beside its card rather than inside it: Trustpilot states that the company has no recent history of asking for those reviews, so they are reported on that profile instead of being averaged into a row. Its composite rose nine points on 18 August 2026 when the record moved out, and the record is still the first thing to read before subscribing. A card was withdrawn from this page on 5 September 2026, and the reason is worth knowing before you trust any capability figure here: Adobe Firefly's rested on an arena entry that has since left the board, no independent replacement was reachable, and the card was pulled rather than published with a hole. Every capability number on this page is a position inside a board that can renormalise or drop an entrant without notice. Its arena figure is the lowest cell on its own card. Ideogram's usability figure was filled on 14 August 2026 from a Trustpilot record that had been sitting unread in its own file after a single G2 review was taken for the whole crowd, and it was re-counted downward on 18 August when those reviews turned out to be entering the sum twice. Those are three different reasons to buy or not buy, and no single number orders them.
Where these numbers come from
Capability is sourced from public output-quality arenas (blind pairwise votes) or from a published accuracy study where no arena covers the category, usability from review-crowd aggregates weighted by sample size, value from verified pricing, and commercial terms from clause positions read off the vendor’s own legal documents on a stated date. Where none of those exists for a tool, the row carries a named criterion instead: a different measurement, taken from saved sources under its own rubric, and the card prints that criterion’s name and the question it answers in place of the axis heading, so the row is never read as the axis it could not fill. Full detail: methodology.
Nothing on this page is placed by hand: the order is the composite, recomputed every time the site is built, and the badge goes to whichever card holds the largest one that day. What the order cannot tell you is how much evidence each number rests on. Krea's usability cell converts 148 reviews and Leonardo AI's converts 48; Recraft's converts 492. A number resting on 48 readings and a number resting on 633 are both honest and they are not the same kind of evidence, which is why the sample sizes are printed on every profile and the measured-cell count sits beside every score. On 18 August 2026 the usability cells for Midjourney and Leonardo AI stopped averaging two consumer records together, because the platforms disagree about how each sample was gathered; where that happened the withheld record is printed on the tool's own page with its own count and date, and the cell left behind rests on fewer readings than the one it replaced. Midjourney rose on that change and Leonardo AI fell, which is what a rule that cannot see which way a record points looks like from outside. Ideogram also fell, for an unrelated reason: one record it had been counting twice.
On commercial ties, as of August 2026: we have joined no affiliate programme covering image generators, so nothing on this page is earning us a commission. If that changes, this line changes with it.
One caution about the value cells, because it is ours rather than a vendor's. Every value cell here was rebuilt on 11 August 2026 onto a single billing basis, the monthly rate, after a browser re-read found cells scored from a billing view a static capture cannot switch. The monthly basis was chosen because Adobe publishes no annual Firefly rate at all, so an annual basis would have required inventing a figure. Two of those re-reads were not label corrections but real errors, and both are recorded on the profiles that carried them.
Questions buyers actually ask
How should I read the score on an AI image generator card?+
As a measurement of that tool and nothing else. A composite is the geometric mean of the dimensions we could fill for that tool, and cards differ in how much stands behind each one: Ideogram's usability cell rests on 162 reviews from one platform, while Recraft's pools 517 across three. This page is ordered by composite, largest first, and that order is recomputed every time the site is built rather than typed in by anyone. What the order cannot tell you is that two composites built from different cells are not two placings in a race, so read the dimensions and the sample sizes printed beside them on each profile before you read the position.
Why is Midjourney's composite the lowest in this category when people like the images so much?+
Because the composite carries more than the pictures, and each half is measured separately. Midjourney's usability cell reads 87.8 and rests on 191 favourable readings alone, converting a raw 90.6: the 366 reviews on Trustpilot are not averaged into it, because that profile is not gathered the way the other platforms are, and they are reported on the profile page with their own date and count instead. Its capability cell, 71.6, converts an arena result read 5 September 2026, and its value and contract cells are both in the eighties. Nothing sits outside the table adjusting the total: the four rows are combined by multiplication, which is why one weak row costs more than it would in an average. Read the rows and the sample sizes printed beside them on the profile before you read the position.
What happened to the Adobe Firefly card?+
A card was withdrawn from this page on 5 September 2026 and the reason answers this question better than the card did. Adobe Firefly's capability cell read an arena entry that has since been removed from the board; no independent measurement of its output was reachable to replace it, and a substitute built from what vendors publish about their entry tiers was measured and refused, because no component of it reaches four of six vendors in a comparable unit. The card was pulled rather than published with an empty cell. Its evidence stays on disk and the page 301s here.
Is any AI image generator free tier usable for commercial work?+
Read each free tier's own terms, because they differ and the differences are the whole answer. Recraft's plan page states that free-plan images are public, that there is no commercial use, and its FAQ says those images are owned by Recraft. Ideogram's free plan runs on 10 slow credits a week with one generation at a time, and its terms state no per-tier commercial rule at all. Leonardo AI's section 8.7 is the sharpest of them: ownership of a free subscriber's output vests in Leonardo on creation and the free user assigns their interest in it, while section 8.3 vests it in paid subscribers. Midjourney publishes no free tier; its cheapest row is a paid plan. Plans read 11 August 2026, terms 12 August 2026.
