1
Midjourney V8.2
Midjourney · Art direction, mood and a distinctive look tuned to your taste
$$
Midjourney is still the creator consensus pick for taste and artistic range, and V8.2 (default since 24 July 2026) was billed as an aesthetics and personalisation update with fewer weak outputs. It has no public API, so it does not appear on LMArena or Artificial Analysis; its rank here rests on creator consensus and its unmatched style tooling, not arena scores.
- Cheapest way in
- No free tier; Basic from $10/month ($8 billed yearly)
- Cost per output
- No per-image price; a standard job uses about 0.8 GPU minutes (1.3 with --hd)
- Style references
- Yes (--sref)
- Personalisation
- Yes, ranking-based profiles (--p)
- Max resolution
- Native 2K (--hd)
- Text rendering
- Short phrases, roughly 1 to 4 words
- Open weights
- No
- API
- No
Good at
- Strongest painterly, cinematic and editorial looks with little prompting
- Style references (--sref) and personalisation profiles (--p) let you lock a consistent house style
- Native 2K generation, plus editor, Vary, Zoom and video tools in one app
Watch out for
- No official public API and no open weights
- Text rendering limited to short strings
- No free trial; not independently benchmarked
Prices tech-insider.org, checked 1 November 2026
Midjourney V8.2 website Work made with it on AI LIVE
2
Reve 2.1
Reve · Polished, art-directed images with reliable layout and lettering
$$
Reve 2.1 is the highest-ranked independent lab model on LMArena's text-to-image board (#7, behind only the big labs), and its plan-then-render approach gives clean compositions at native 4K. It pairs strong aesthetics with legible text, which makes it a practical style tool for posters and key art.
- Cheapest way in
- Free tier with daily energy; Lite $7.99/month, Pro $19.99/month
- Cost per output
- About $0.20 per v2 image or edit via API (150 credits; $10 buys 7,500 credits); also on fal
- Style references
- Not documented
- Personalisation
- No
- Max resolution
- Native 4K (vendor claim)
- Text rendering
- Strong, including small branding text
- Open weights
- No
Good at
- #7 on LMArena Text-to-Image, top independent lab
- Native 4K output (vendor claim)
- Layout-first generation with element-level re-prompting
- Good text in posters and packaging
Watch out for
- No documented style-reference or moodboard feature
- Closed weights
- Arena score still marked preliminary; not on Artificial Analysis for 2.x
Benchmarks LMArena (arena.ai) Text-to-Image Arena: #7, score 1300 (preliminary) · LMArena (arena.ai) Image Edit Arena, single image: #18, score 1373
Prices eesel.ai, checked 1 November 2026
Reve 2.1 website
3
Grok Imagine Image 2.0
SpaceXAI · Striking, high-impact images at a low per-image price
$
Grok Imagine Image 2.0 sits in the top six on both major text-to-image arenas (#6 LMArena, #6 Artificial Analysis, tied 4th to 6th), ahead of every other model on this list that has been benchmarked. At $0.04 per image via API it is one of the cheapest ways to get frontier-level looks.
- Cheapest way in
- In the Grok app; $0.04 per image via the xAI API
- Cost per output
- $0.04 per image (grok-imagine-image-2.0) via xAI API; resellers list higher 2K rates
- Style references
- Not documented
- Personalisation
- No
- Max resolution
- Up to 2K via resellers; xAI does not state it
- Text rendering
- Not documented
- Open weights
- No
Good at
- Top-six text-to-image on both arenas
- $0.04 per image official API price
- Same model handles editing
Watch out for
- No style-reference or moodboard tooling documented
- Arena score on LMArena still preliminary
- Closed weights
Benchmarks LMArena (arena.ai) Text-to-Image Arena: #6, score 1317 (preliminary) · Artificial Analysis Text to Image Arena (AA-Image-T2I v2.1): #6 (tied 4-6), Elo 1166
Prices docs.x.ai, checked 1 November 2026
Grok Imagine Image 2.0 website
4
Seedream 5.0 Pro
ByteDance Seed · Photographic and illustrative looks with multi-reference blending
$$
Seedream 5.0 Pro is a well-tested all-rounder (over 100,000 LMArena votes) at #14 on LMArena text-to-image and #16 on Artificial Analysis. It is widely available inside creator platforms and also ranks highly for editing, so it earns a place in both lists.
- Cheapest way in
- Pay per use via BytePlus ModelArk, resellers and platforms like Artlist
- Cost per output
- About $0.045 to $0.09 per image depending on provider (Artificial Analysis lists $0.09)
- Style references
- Yes, via reference images
- Max resolution
- 2K confirmed; 4K disputed
- Text rendering
- Multilingual (14 to 15 languages reported)
- Open weights
- No
- Reference images
- Up to 14 (OpenRouter) or 10 (other sources)
Good at
- Large vote base, so its rankings are stable
- Many reference images per request for style and subject blending
- Text in many languages
Watch out for
- Sources disagree on 4K support; 2K is the confirmed tier
- Some creators rate its look below Seedream 4.5 for certain styles
- Closed weights
Benchmarks LMArena (arena.ai) Text-to-Image Arena: #14, score 1250 · Artificial Analysis Text to Image Arena (AA-Image-T2I v2.1): #16, Elo 1100
Prices atlascloud.ai, checked 1 November 2026
Seedream 5.0 Pro website Work made with it on AI LIVE
5
FLUX 3 Image
Black Forest Labs · Pro-grade realism plus an open-weights option you can fine-tune
$
FLUX 3 Image ranks #13 on Artificial Analysis text-to-image, and the 5.6B open-weights Klein version (15 October 2026) generates natively at 4K with up to 10 references. The FLUX family has the deepest LoRA and ComfyUI ecosystem, which is what lets creators build and own a custom style.
- Cheapest way in
- Pay as you go, no subscription; Klein weights downloadable
- Cost per output
- $0.048 per image (FLUX 3 Image); from $0.009 per 1K image (Klein)
- Style references
- Via multi-reference input (Klein up to 10) and LoRAs
- Personalisation
- Custom LoRA training via ecosystem
- Max resolution
- Up to 4K (native 4K on Klein)
- Text rendering
- Not documented for FLUX 3
- Open weights
- Klein only (FLUX Kommunity License)
Good at
- Klein is open weights, free for commercial use under $5M annual revenue
- Native 4K (Klein)
- Huge fine-tuning and LoRA ecosystem
Watch out for
- Flagship FLUX 3 Image weights are not public
- FLUX 3 not yet on LMArena's text-to-image board
- No built-in moodboard or personalisation app
Benchmarks Artificial Analysis Text to Image Arena (AA-Image-T2I v2.1): FLUX 3 Image #13 (tied 8-15), Elo 1117; Klein #21, Elo 1043 · LMArena (arena.ai) Text-to-Image Arena: FLUX.2 [max] #28, score 1161 (FLUX 3 not yet listed)
Prices bfl.ai, checked 1 November 2026
FLUX 3 Image website
6
Krea 2
Krea · Steering a look with moodboards and style references
$
Krea 2 (May 2026) is built specifically for style control, with image style references, moodboards and a creativity setting, plus open checkpoints on Hugging Face. It trails the leaders on arenas (#24 on Artificial Analysis text-to-image), but no other benchmarked model gives creators this much direct aesthetic steering.
- Cheapest way in
- Free plan (100 compute units/day); Basic from $9/month ($5 billed yearly)
- Cost per output
- API $0.03 (Medium) or $0.06 (Large) per image; +$0.005 with style references, +$0.01 with a moodboard
- Style references
- Yes (up to 4 reported)
- Moodboards
- Yes, 1 per request
- Max resolution
- 1K
- Text rendering
- Not a stated strength
- Open weights
- Partly (RAW and Turbo checkpoints)
Good at
- Style references and moodboards built into the model and API
- Open checkpoints (Krea 2 RAW, Krea 2 Turbo)
- LoRA training on Krea plans, including the free plan
Watch out for
- 1K output only
- Mid-table on both arenas
- Unclear which hosted variants match the open checkpoints
Benchmarks Artificial Analysis Text to Image Arena (AA-Image-T2I v2.1): Krea 2 Large #24, Elo 1034 · LMArena (arena.ai) Text-to-Image Arena: Krea 2 Medium #44, score 1128
Prices krea.ai, checked 1 November 2026
Krea 2 website Work made with it on AI LIVE
7
Ideogram 4.0 / 4.5
Ideogram · Graphic design, posters and anything where the type matters
$$
Ideogram 4.0 (Quality) ranks #20 on LMArena text-to-image and remains the design-led choice for typography-heavy work. Version 4.5 (30 September 2026) adds precise, low-drift editing, so the same tool covers layout and revisions.
- Cheapest way in
- Free tier (30 points a week); Plus $20/month ($15 billed yearly)
- Cost per output
- API $0.03 (Low), $0.06 (Medium), $0.10 (High) per image; $0.22 at High for reference generation or Precise Edit
- Style references
- Reference images supported (up to 4 in Precise Edit)
- Max resolution
- Native 2K
- Text rendering
- Core strength
- Open weights
- 4.0 only, non-commercial licence
Good at
- Typography and layout focus
- Native 2K
- Clear tiered API pricing
Watch out for
- Middling on Artificial Analysis text-to-image (#32 for 4.0 Quality)
- Promised open weights not released
- Reference and edit jobs cost more ($0.22 at High)
Benchmarks LMArena (arena.ai) Text-to-Image Arena: Ideogram 4.0 (Quality) #20, score 1200 · Artificial Analysis Text to Image Arena (AA-Image-T2I v2.1): Ideogram 4.0 (Quality) #32, Elo 1026; 4.5 (High) #40, Elo 1015
Prices glbgpt.com, checked 1 November 2026
Ideogram 4.0 / 4.5 website
8
Recraft V4.1
Recraft · Brand-consistent illustration, icons and editable vector art
$
Recraft V4.1 is the only model here that outputs editable SVG, and it accepts up to 10 style reference images or a saved style ID to keep a brand look. Its Utility Pro variant reaches #26 on LMArena text-to-image; it is a design-system tool more than a mood-piece generator.
- Cheapest way in
- Free plan (no commercial rights); API from $0.035 per image
- Cost per output
- API $0.035 raster, $0.08 vector, $0.21 Pro raster (2K), $0.30 Pro vector; style creation $0.005
- Style references
- Up to 10, or a saved style_id
- Max resolution
- 2K (Pro)
- Text rendering
- Not documented for V4.1
- Open weights
- No
- Vector
- Yes
Good at
- Editable SVG output
- Saved styles and up to 10 style references
- Cheap raster tier ($0.035)
Watch out for
- Max 2K raster
- Free plan images are owned by Recraft and not licensed for commercial use
- Less painterly range than Midjourney
Benchmarks LMArena (arena.ai) Text-to-Image Arena: V4.1 Utility Pro #26, score 1168 (low vote count); V4.1 Pro #42 · Artificial Analysis Text to Image Arena (AA-Image-T2I v2.1): V4.1 Utility #29, Elo 1026
Prices recraft.ai, checked 1 November 2026
Recraft V4.1 website
9
Qwen-Image-2.1
Alibaba · Running a strong-looking model locally for research and experiments
$
Qwen-Image-2.1 (20 September 2026) is the best-ranked open-weights image model on LMArena text-to-image at #19, and outputs native 2048x2048 with RGBA transparency. Its licence changed to a non-commercial research licence, so it suits testing and personal pipelines rather than client work.
- Cheapest way in
- Free to download and run locally (research licence)
- Cost per output
- No hosted API; the closed Qwen-Image-3.0 costs about $0.03-0.04 per image
- Style references
- Via community tooling
- Max resolution
- 2048x2048
- Text rendering
- Qwen line known for text; not separately verified for 2.1
- Open weights
- Yes, non-commercial research licence
- Parameters
- 7B DiT
Good at
- Open weights, runs in ComfyUI
- Native 2K with transparency (RGBA)
- Highest open-weights score on LMArena text-to-image
Watch out for
- Non-commercial Qwen Research License
- No hosted API listed on Artificial Analysis
- Successor Qwen-Image-3.0 is closed
Benchmarks LMArena (arena.ai) Text-to-Image Arena: #19, score 1225 (preliminary) · Artificial Analysis Text to Image Arena (AA-Image-T2I v2.1): #22, Elo 1042
Prices aiweekly.co, checked 1 November 2026
Qwen-Image-2.1 website
