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

Best image editing models to run locally on a 16GB graphics card

A leaderboard of the best image editing models you can run locally on a 16GB graphics card (with 32GB of system RAM). Qwen-Image-2.1 is #1 right now, followed by Qwen-Image-Edit-2509 and Qwen-Image-Edit-2511. Models are ranked on human votes and image editing tests. Each model shows the download that fits and the command to run it.

18 models · scores combine 15 benchmarks · updated how it works · find a leaderboard · benchmark your machine
task[all tasks][text to image][image to image / editing]

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

Models for 32GB RAM · 16GB graphics card → ~16GB fast · local only · reset
+ more
card[RTX 4060 Ti 16GB][RTX 4070 Ti Super][RTX 4080][RTX 4080 Super][RTX 5060 Ti 16GB][RTX 5070 Ti][RTX 5080][RTX 3080 Laptop][RTX 3080 Ti Laptop][RTX 4090 Laptop][RTX 5080 Laptop][RTX A4000][RX 7600 XT][RX 7800 XT][RX 7900 GRE][RX 9060 XT][RX 9070][RX 9070 XT]

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=32&gpu=card&vram=16&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]
vram[1][2][4][8][12][16][24][32][48][64][80][96][128][160]GB
price[local][$0.002][$0.005][$0.01][$0.02][$0.05][$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
15Qwen-Image-2.1 · gguf-q4_k + Qwen3-VL 8B q4_k_m (sd.cpp) Alibaba66±3*local~12GBhow to run › LMArena leaderboard dataset
36Qwen-Image-Edit-2509 · gguf-q4_k_s + Qwen2.5-VL q8_0 (sd.cpp) Alibaba51±3local (slow)~26GBhow to run › Meituan LongCat — LongCat-Image technical reportGEditBench v2: A Human-Aligned Benchmark for General Image EditingBAAI — OmniGen2 technical report
38Qwen-Image-Edit-2511 · gguf-q4_k_m + Qwen2.5-VL bf16 (sd.cpp) Alibaba51±2local (slow)~37GBhow to run › Edit-Compass & EditReward-Compass: A Unified Benchmark for Image Editing and Reward ModelingLMArena leaderboard datasetGEditBench v2: A Human-Aligned Benchmark for General Image EditingMultiTextEdit: Benchmarking Cross-Lingual Degradation in Text-in-Image Editing (arXiv:2605.08163v2), Table 3
39FLUX.2 [klein] 9B · gguf-q4_0 + Qwen3 8B q4_k_m (sd.cpp) Black Forest Labs51±3*local~14GBhow to run › LMArena leaderboard datasetGEditBench v2: A Human-Aligned Benchmark for General Image Editing
41LongCat-Image-Edit · bf16 + model CPU offload Meituan50±2local (slow)~36GBhow to run › Edit-Compass & EditReward-Compass: A Unified Benchmark for Image Editing and Reward ModelingMeituan LongCat — LongCat-Image technical reportGEditBench v2: A Human-Aligned Benchmark for General Image Editing
47Qwen-Image-Edit · gguf-q4_k_m + Qwen2.5-VL q8_0 (sd.cpp) Alibaba49±2local (slow)~27GBhow to run › Meituan LongCat — LongCat-Image technical reportLMArena leaderboard datasetGEditBench v2: A Human-Aligned Benchmark for General Image EditingUnicEdit-10M: A Dataset and Benchmark Breaking the Scale-Quality Barrier via Unified Verification for Reasoning-Enriched EditsGEBench: Benchmarking Image Generation Models as GUI EnvironmentsMultiTextEdit: Benchmarking Cross-Lingual Degradation in Text-in-Image Editing (arXiv:2605.08163v2), Table 3
48FLUX.2 [dev] · nf4 (bnb-4bit DiT + text encoder) Black Forest Labs49±2local (slow)~20GBhow to run › Edit-Compass & EditReward-Compass: A Unified Benchmark for Image Editing and Reward ModelingLMArena leaderboard datasetGEditBench v2: A Human-Aligned Benchmark for General Image Editing
51FLUX.2 [klein] 4B · bf16 (cpu offload) Black Forest Labs47±3*local~10GBhow to run › LMArena leaderboard datasetGEditBench v2: A Human-Aligned Benchmark for General Image Editing
56Qwen-Image · nf4 (bitsandbytes 4-bit transformer + text encoder) Alibaba46±5*local~13GBhow to run › ServImage: An Image Generation and Editing Benchmark from Real-world Commercial Imaging Services (arXiv:2604.24023v2)
65FLUX.1 Kontext [dev] · gguf-q8_0 (sd.cpp) Black Forest Labs36±2local (slow)~28GBhow to run › Meituan LongCat — LongCat-Image technical reportLMArena leaderboard datasetEdit-Compass & EditReward-Compass: A Unified Benchmark for Image Editing and Reward ModelingUnicEdit-10M: A Dataset and Benchmark Breaking the Scale-Quality Barrier via Unified Verification for Reasoning-Enriched EditsGEditBench v2: A Human-Aligned Benchmark for General Image EditingCREval: An Automated Interpretable Evaluation for Creative Image Manipulation under Complex Instructions (arXiv:2603.26174v1)+2
67ChronoEdit-14B · bf16 + model offload (official script) NVIDIA34±3*local (slow)~34GBhow to run › Edit-Compass & EditReward-Compass: A Unified Benchmark for Image Editing and Reward Modeling
69OmniGen2 · bf16 (official repo) BAAI34±2local (slow)~17GBhow to run › Edit-Compass & EditReward-Compass: A Unified Benchmark for Image Editing and Reward ModelingUnicEdit-10M: A Dataset and Benchmark Breaking the Scale-Quality Barrier via Unified Verification for Reasoning-Enriched EditsGEditBench v2: A Human-Aligned Benchmark for General Image EditingCREval: An Automated Interpretable Evaluation for Creative Image Manipulation under Complex Instructions (arXiv:2603.26174v1)BAAI — OmniGen2 technical reportMultiTextEdit: Benchmarking Cross-Lingual Degradation in Text-in-Image Editing (arXiv:2605.08163v2), Table 3+1
70BAGEL · nf4 (app.py --mode 2, bitsandbytes) ByteDance Seed33±2local~12GBhow to run › LMArena leaderboard datasetEdit-Compass & EditReward-Compass: A Unified Benchmark for Image Editing and Reward ModelingUnicEdit-10M: A Dataset and Benchmark Breaking the Scale-Quality Barrier via Unified Verification for Reasoning-Enriched EditsGEditBench v2: A Human-Aligned Benchmark for General Image EditingCREval: An Automated Interpretable Evaluation for Creative Image Manipulation under Complex Instructions (arXiv:2603.26174v1)BAAI — OmniGen2 technical report+3
71Stable Diffusion 3.5 Large · fp8 (TensorRT) Stability AI33±7*local~5.4GBhow to run › ServImage: An Image Generation and Editing Benchmark from Real-world Commercial Imaging Services (arXiv:2604.24023v2)
72UNO · fp8 FLUX.1-dev base + UNO LoRA, CPU offload ByteDance32±7*local (slow)~16GBhow to run › BAAI — OmniGen2 technical report
74OmniGen (v1) · official OmniGen package, default settings BAAI30±7*†local~9GBhow to run › BAAI — OmniGen2 technical report
80OneCAT-3B · bf16 (official generate_txt2img.py) Meituan (OneCAT team)14±5*local (slow)~25GBhow to run › Edit-Compass & EditReward-Compass: A Unified Benchmark for Image Editing and Reward Modeling
81InstructPix2Pix · fp16 UC Berkeley (Brooks, Holynski, Efros)10±4local~2.6GBhow to run › UnicEdit-10M: A Dataset and Benchmark Breaking the Scale-Quality Barrier via Unified Verification for Reasoning-Enriched EditsEdit-Compass & EditReward-Compass: A Unified Benchmark for Image Editing and Reward ModelingImagenWorld: Stress-Testing Image Generation Models with Explainable Human Evaluation on Open-ended Real-World TasksMultiTextEdit: Benchmarking Cross-Lingual Degradation in Text-in-Image Editing (arXiv:2605.08163v2), Table 3

+ 63 more don't fit your machine or budget · show them

81 of 81 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