whatmodel.sh · find the best image models you can run

Best text-to-image models to run locally on a 16GB Mac

A leaderboard of the best text-to-image models you can run locally on a Mac with 16GB of memory. Ideogram 4.0 is #1 right now, followed by Z-Image (Base) and Z-Image-Turbo. Models are ranked on human votes and image generation tests like GenEval. Each model shows the download that fits and the command to run it.

24 models · scores combine 41 benchmarks · updated how it works · find a leaderboard · benchmark your machine
task[all tasks][text to image][image to image / editing]

Scores use reviewed text to image evaluations only. Compare scores within this task. Models without task measurements are excluded.

Models for 16GB RAM · Apple Silicon → ~11GB fast · local only · reset
+ more
chip[M1][M1 Pro][M2][M2 Pro][M3][M4][M5][M6 (16GB)]
set[all sets][LMArena text-to-image · 157][CVTG-2K word accuracy · 3]

Filter the model list while keeping task scores. Models without a shared score scale are not yet calibrated; select a set to see its own scores.

Detect your exact chip or graphics card:

curl -G whatmodel.sh/image -d max=0 --data-urlencode "mac=$(sysctl -n hw.memsize machdep.cpu.brand_string)"Mac

curl -G whatmodel.sh/image -d max=0 --data-urlencode "ram=$(awk '/MemT/{print $2}' /proc/meminfo)" --data-urlencode "nv=$(nvidia-smi --query-gpu=name,memory.total --format=csv,noheader 2>/dev/null)"Linux

curl.exe -G whatmodel.sh/image -d max=0 --data-urlencode "ram=$((gcim Win32_ComputerSystem).TotalPhysicalMemory)" --data-urlencode "nv=$(nvidia-smi --query-gpu=name,memory.total --format=csv,noheader 2>$null)"Windows

$ terminal

detecting your OS…

Run curl "whatmodel.sh/image?ram=16&gpu=apple&max=0"to get this list in your terminal (add &format=json for agents).

ram[8][16][24][32][48][64][96][128][192][256]GB
gpu[apple silicon][graphics card][none]
price[local][$0.001][$0.005][$0.02][$0.1][any]max per image

detecting your OS…

Score vs memory to run

#modeltask score↓where↓speed↓RAM↓get itsources
#modeltask score↓where↓speed↓RAM↓get itsources
18Ideogram 4.0 · mlx-q4 Ideogram68±2*local~6.1GBhow to run › LMArena leaderboard dataset
53Z-Image (Base) · gguf-q4_k_m + Qwen3 4B q4_k_m (sd.cpp) Alibaba (Tongyi-MAI)58±2local~10GBhow to run › Tongyi-MAI (Alibaba) — Z-Image technical reportUltraText Bench: A Comprehensive Bilingual Benchmark for Evaluating Visual Text Rendering in Image Generation (arXiv:2610.09823v1)QC-T2I-Bench: Scalable Question-Centric Text-to-Image Evaluation (Reliable Ranking, Fine-Grained Diagnosis, and Cost-Aware Routing)
62Z-Image-Turbo · gguf-q3_k (+ Qwen3-4B Q4_K_M text encoder) Alibaba (Tongyi-MAI)56±1local~7.7GBhow to run › LMArena leaderboard datasetTongyi-MAI (Alibaba) — Z-Image technical reportUltraText Bench: A Comprehensive Bilingual Benchmark for Evaluating Visual Text Rendering in Image Generation (arXiv:2610.09823v1)QC-T2I-Bench: Scalable Question-Centric Text-to-Image Evaluation (Reliable Ranking, Fine-Grained Diagnosis, and Cost-Aware Routing)Can Image Models Imagine Time? ImageTime: A Novel Benchmark for Probing Visual World Modeling Through Spatiotemporal ConsistencyFaithful, Enriched, and Precise: Benchmarking Natural-Science Illustration Generation by T2I models (FEPBench)+1
78FLUX.2 [klein] 4B · gguf-q4_0 + Qwen3 4B q4_k_m (sd.cpp) Black Forest Labs49±2local~6.9GBhow to run › LMArena leaderboard datasetQC-T2I-Bench: Scalable Question-Centric Text-to-Image Evaluation (Reliable Ranking, Fine-Grained Diagnosis, and Cost-Aware Routing)UltraText Bench: A Comprehensive Bilingual Benchmark for Evaluating Visual Text Rendering in Image Generation (arXiv:2610.09823v1)
97FLUX.1 [dev] · mlx-4bit Black Forest Labs43±1local~7.7GBhow to run › LMArena leaderboard datasetT2I-ReasonBench: Benchmarking Reasoning-Informed Text-to-Image Generation (Findings of ACL 2026)TIIF-Bench: How Does Your T2I Model Follow Your Instructions? (arXiv v3)QC-T2I-Bench: Scalable Question-Centric Text-to-Image Evaluation (Reliable Ranking, Fine-Grained Diagnosis, and Cost-Aware Routing)MMMG: A Massive, Multidisciplinary, Multi-Tier Generation Benchmark for Text-to-Image Reasoning (arXiv:2506.10963v2)GenExam leaderboard (OpenGVLab/GenExam README, relaxed score)+3
98Playground v2 (1024px aesthetic) · fp16 (sd.cpp) Playground43±4*local~8.8GBhow to run › GenAI Arena: An Open Evaluation Platform for Generative Models (arXiv v3)
101FLUX.1 [schnell] · mlx-4bit Black Forest Labs42±3*local~7.7GBhow to run › T2I-ReasonBench: Benchmarking Reasoning-Informed Text-to-Image Generation (Findings of ACL 2026)WISE: A World Knowledge-Informed Semantic Evaluation for Text-to-Image Generation (arXiv v4)
113SDXL-Lightning (4-step UNet) · fp16 (sd.cpp) ByteDance40±4*local~8.8GBhow to run › GenAI Arena: An Open Evaluation Platform for Generative Models (arXiv v3)
114Stable Diffusion 3 Medium · q8_0 (sd.cpp --type) Stability AI39±2local~11GBhow to run › T2I-ReasonBench: Benchmarking Reasoning-Informed Text-to-Image Generation (Findings of ACL 2026)EvalMuse-40K: A Reliable and Fine-Grained Benchmark with Comprehensive Human Annotations for Text-to-Image Generation Model Evaluation (arXiv v2)WISE: A World Knowledge-Informed Semantic Evaluation for Text-to-Image Generation (arXiv v4)HPSv3 repository README, image generation benchmark table (MizzenAI/HPSv3)
122SSD-1B · fp16 (sd.cpp) Segmind36±3*local~5.9GBhow to run › EvalMuse-40K: A Reliable and Fine-Grained Benchmark with Comprehensive Human Annotations for Text-to-Image Generation Model Evaluation (arXiv v2)
127Stable Diffusion XL 0.9 (base) · q8_0 (sd.cpp --type) Stability AI35±3*local~4.9GBhow to run › tgxs002/HPSv2 repository README, HPS v2 benchmark tableWISE: A World Knowledge-Informed Semantic Evaluation for Text-to-Image Generation (arXiv v4)
128SDXL-Turbo · fp16 (sd.cpp) Stability AI35±2local~8.8GBhow to run › EvalMuse-40K: A Reliable and Fine-Grained Benchmark with Comprehensive Human Annotations for Text-to-Image Generation Model Evaluation (arXiv v2)GenAI Arena: An Open Evaluation Platform for Generative Models (arXiv v3)ConceptMix: A Compositional Image Generation Benchmark with Controllable Difficulty (arXiv v1)
129Stable Diffusion XL 1.0 · fp16 (sd.cpp) Stability AI35±1local~8.8GBhow to run › EvalMuse-40K: A Reliable and Fine-Grained Benchmark with Comprehensive Human Annotations for Text-to-Image Generation Model Evaluation (arXiv v2)GenAI Arena: An Open Evaluation Platform for Generative Models (arXiv v3)GenEval authors' results: djghosh13/geneval README and arXiv 2310.11513v1ImagenWorld: Stress-Testing Image Generation Models with Explainable Human Evaluation on Open-ended Real-World TasksConceptMix: A Compositional Image Generation Benchmark with Controllable Difficulty (arXiv v1)Kandinsky 3.0 — original human comparisons+5
133Dreamlike Photoreal 2.0 · q8_0 (sd.cpp --type) Dreamlike.art34±4*local~2.1GBhow to run › tgxs002/HPSv2 repository README, HPS v2 benchmark table
146DALL·E Mega (minDALL-E) · fp32 DALL·E Mini / minDALL-E31±3*local~8GBhow to run › tgxs002/HPSv2 repository README, HPS v2 benchmark tableTIFA: Accurate and Interpretable Text-to-Image Faithfulness Evaluation with Question Answering (arXiv v3)
149Openjourney · q8_0 (sd.cpp --type) PromptHero29±2local~2.1GBhow to run › GenAI Arena: An Open Evaluation Platform for Generative Models (arXiv v3)tgxs002/HPSv2 repository README, HPS v2 benchmark tableImagenHub: Standardizing the evaluation of conditional image generation models (arXiv v4)
150Stable Diffusion v1.4 · fp32 CompVis28±2local~8GBhow to run › GenEval authors' results: djghosh13/geneval README and arXiv 2310.11513v1tgxs002/HPSv2 repository README, HPS v2 benchmark tableConceptMix: A Compositional Image Generation Benchmark with Controllable Difficulty (arXiv v1)
151Stable Diffusion v1.3 · fp32 CompVis28±4*local~8GBhow to run › GenEval authors' results: djghosh13/geneval README and arXiv 2310.11513v1
152Stable Diffusion v1.1 · fp32 CompVis27±3local~8GBhow to run › GenEval authors' results: djghosh13/geneval README and arXiv 2310.11513v1TIFA: Accurate and Interpretable Text-to-Image Faithfulness Evaluation with Question Answering (arXiv v3)REAL: Realism Evaluation of Text-to-Image Generation Models for Effective Data Augmentation (arXiv:2502.10663v1)
153Stable Diffusion v1.5 · fp32 Stability AI27±2local~8GBhow to run › EvalMuse-40K: A Reliable and Fine-Grained Benchmark with Comprehensive Human Annotations for Text-to-Image Generation Model Evaluation (arXiv v2)GenEval authors' results: djghosh13/geneval README and arXiv 2310.11513v1TIFA: Accurate and Interpretable Text-to-Image Faithfulness Evaluation with Question Answering (arXiv v3)Evaluating Numerical Reasoning in Text-to-Image Models (GeckoNum, arXiv v3)WISE: A World Knowledge-Informed Semantic Evaluation for Text-to-Image Generation (arXiv v4)
154Stable Diffusion v1.2 · fp32 CompVis27±3*local~8GBhow to run › EvalMuse-40K: A Reliable and Fine-Grained Benchmark with Comprehensive Human Annotations for Text-to-Image Generation Model Evaluation (arXiv v2)GenEval authors' results: djghosh13/geneval README and arXiv 2310.11513v1
155LCM Dreamshaper v7 · openvino-fp16 (OpenVINO GenAI) SimianLuo (LCM authors)24±4*local~3GBhow to run › GenAI Arena: An Open Evaluation Platform for Generative Models (arXiv v3)
—Mage-Flow · int8 convrot + Qwen3-VL 4B q4_k_m (sd.cpp) Microsoftnot yet calibratedlocal~8.9GBhow to run › Mage-Flow: An Efficient Native-Resolution Foundation Model for Image Generation and Editing
—Mage-Flow-Turbo · int8 convrot + Qwen3-VL 4B q4_k_m (sd.cpp) Microsoftnot yet calibratedlocal~8.9GBhow to run › Mage-Flow: An Efficient Native-Resolution Foundation Model for Image Generation and Editing

+ 138 more don't fit your machine or budget · show them

162 of 162 catalogue models listed here · sources and licences

* new or thinly tested: ranked on too few benchmark results · † only the maker’s own numbers so far · ~RAM is estimated; full context may need more memory